L-serine synthesis via the phosphorylated pathway in humans

Abstract

L-serine is a nonessential amino acid in eukaryotic cells, used for protein synthesis and in producing phosphoglycerides, glycerides, sphingolipids, phosphatidylserine, and methylenetetrahydrofolate. Moreover, L-serine is the precursor of two relevant coagonists of NMDA receptors: glycine (through the enzyme serine hydroxymethyltransferase), which preferentially acts on extrasynaptic receptors and D-serine (through the enzyme serine racemase), dominant at synaptic receptors. The cytosolic “phosphorylated pathway” regulates de novo biosynthesis of L-serine, employing 3-phosphoglycerate generated by glycolysis and the enzymes 3-phosphoglycerate dehydrogenase, phosphoserine aminotransferase, and phosphoserine phosphatase (the latter representing the irreversible step). In the human brain, L-serine is primarily found in glial cells and is supplied to neurons for D-serine synthesis. Serine-deficient patients show severe neurological symptoms, including congenital microcephaly, psychomotor retardation, and intractable seizures, thus highlighting the relevance of de novo production of this amino acid in brain development and morphogenesis. Indeed, the phosphorylated pathway is strictly linked to cancer. Moreover, L-serine has been suggested as a ready-to-use treatment, as also recently proposed for Alzheimer’s disease. Here, we present our current state of knowledge concerning the three mammalian enzymes of the phosphorylated pathway and known mutations related to pathological conditions: although the structure of these enzymes has been solved, how enzyme activity is regulated remains largely unknown. We believe that an in-depth investigation of these enzymes is crucial to identify the molecular mechanisms involved in modulating concentrations of the serine enantiomers and for studying the interplay between glial and neuronal cells and also to determine the most suitable therapeutic approach for various diseases.

Introduction

L-serine plays a versatile role in intermediary metabolism in eukaryotic cells, serving as a building block for protein synthesis and as a precursor for molecules that are essential for cell proliferation, growth, differentiation, and function: phosphoglycerides, glycerides [1], complex macromolecules such as sphingolipids and phosphatidylserine [2], as well as methylenetetrahydrofolate, which is relevant as a source of one-carbon units for methylation processes and nucleotide synthesis [3]. Moreover, L-serine is the precursor of two neuroactive signaling molecules: glycine and D-serine. L-serine is converted to glycine through the reversible reaction catalyzed by the enzyme serine hydroxymethyltransferase (SHMT, EC 2.1.2.1) [4] and to D-serine by the reversible reaction due to serine racemase (SR, EC 5.1.1.18) [5]; see Fig. 1. L-serine metabolism plays a major role in the development and function of the central nervous system (CNS) [6, 7]. This is made apparent by the fact that patients showing a strong decrease in L-serine and glycine levels in plasma/cerebrospinal fluid are affected by severe neurological disorders [8, 9].

Fig. 1.

Schematic model of the L-serine/D-serine pathway in the CNS. Glucose is converted to L-serine via glycolysis and phosphorylated pathways in astrocytes. L-Serine is shuttled to neurons via still unidentified neutral amino acid transporters (most likely ASCT types). Neuronal uptake of L-serine fuels the synthesis of D-serine by neuronal SR. D-serine is released by neurons by the Asc-1 transporter and is subsequently transferred and stored in astrocytes. Here, D-serine cellular concentration is modulated by the catabolic activity of DAAO [27, 28]. The neuromodulator can be released from astrocytes by a transporter (probably Asc-1), Cx43 hemichannels, and an activity-dependent vesicular process, to regulate NMDARs [26, 31, 34]. L-Serine can be reversibly converted into glycine by SHMT, in this way affecting the tetrahydrofolate pool. Notably, D-serine is the main coagonist for synaptic NMDARs, and glycine is the dominant coagonist at extracellular NMDARs [20–22]

In humans, L-serine is available from different sources: from intestinal absorption of dietary proteins, degradation of proteins and phospholipids, conversion of glycine via SHMT, and through the “phosphorylated pathway”. This cytosolic pathway starts from 3-phosphoglycerate generated by glycolysis. It is irreversible and involves three enzymes that are coordinately expressed in many tissues (i.e., brain, kidney, liver, testis, and spleen): 3-phosphoglycerate dehydrogenase (PHGDH, EC 1.1.1.95), phosphoserine aminotransferase (PSAT, EC 2.6.1.52), and phosphoserine phosphatase (PSP, EC 3.1.3.3, which catalyzes the irreversible step). Analysis performed in the framework of metabolic control theory suggested that, under high biosynthetic flux, the pathway is controlled by the last step, while, at low biosynthetic flux, such control might be distributed among the three enzymes [10].

In this review, we will present recent knowledge about the structure–function relationships of the human enzymes involved in the phosphorylated pathway and the genetic defects associated with these enzymes. In-depth knowledge is of utmost relevance, since the investigated enzymes might represent potential targets for various treatments.

An overview on L-serine interplay

The phosphorylated pathway represents the main de novo synthetic pathway of L-serine [11], particularly in the CNS given the low permeability of the blood–brain barrier and the effect and phenotype of genetic defects in the enzymes of this pathway [12, 13]. In the brain, L-serine is mainly produced in astrocytes [14, 15], indicating that it represents an essential amino acid for neurons. PHGDH is abundantly expressed in astrocytes, while cellular levels are barely detectable in neurons, whereas PSAT and PSP are expressed by both cell types [14–16]. There is a strict correlation between L-serine and neurotransmission as it represents the key rate-limiting factor for maintaining steady-state levels of D-serine in the adult brain: mice with a conditional deletion of PHGDH showed that in mature neuronal circuits, L-serine availability determines the rate of D-serine synthesis in the forebrain and controls N-methyl-d-aspartate receptor (NMDAR) function, at least in the hippocampus [12]. NMDARs are unique in that they require the binding of two agonists to be fully functional: glutamate to the GluN2 subunit and a coagonist to the “glycine site” of the GluN1 subunit [17–19]. While the synaptic NMDARs are preferentially activated by the coagonist D-serine, the extrasynaptic ones better interact with the coagonist glycine [20, 21]: the regionalized availability of the coagonist matches the preferential affinity of synaptic receptors for D-serine and of the extrasynaptic NMDARs for glycine. Glycine and D-serine relative availability at rat hippocampal glutamatergic synapses regulate the trafficking and synaptic content of NMDAR subtypes [22]: D-serine alters the membrane dynamics and synaptic content of GluN2B subunit of NMDAR (but not of GluN2A ones) through a process requiring PDZ-binding scaffold partners. Electrophysiological experiments demonstrated that D-serine is required for the induction and expression of Long-Term Potentiation at both excitatory and inhibitory synapses, while glycine does not modulate synaptic plasticity, but controls neuronal gain activity at the dendritic integration level [23]. Indeed, the identity of the NMDAR coagonist at synapses in the lateral nucleus of the amygdala depends on the level of synaptic activation [24].

Glycine generated by SHMT from L-serine (Fig. 1) is involved in both inhibitory neurotransmission via glycine receptors in brain stem and spinal cord and excitatory neurotransmission through activation of NMDARs [17]. Indeed, L-serine can be converted by the pyridoxal-5′-phosphate (PLP)-dependent enzyme SR into the neuromodulator D-serine [5, 25]. D-Serine is then degraded by both the α and β-elimination reaction catalyzed by SR and oxidative deamination catalyzed by d-amino acid oxidase (DAAO, EC 1.4.3.3) [25–28] (Fig. 1). SR is mainly expressed in neurons (although the SR expression in astrocytes is still debated) [21, 29–31], which release D-serine upon membrane depolarization [32], while astrocytes contain the highest amount of D-serine in most brain regions and release this neuromodulator in an activity-dependent manner [33]. Interestingly, L-serine and its precursors are also primarily found in glial cells, suggesting that, although neurons possess high levels of SR, they require an external source of L-serine. Taken together, these observations support the “serine shuttle model”: D-serine synthesized in neurons is shuttled to astrocytes where it is stored and released [34]. Very recently, Wolosker’s group provided further support for this glia-neuron crosstalk, demonstrating that inhibition of astrocytic PHGDH suppressed the de novo synthesis of l- and d-serine and reduced the NMDAR synaptic potentials and long-term potentiation at the Schaffer collateral-CA1 synapse [35]. D-serine has emerged as an influential player in the context of psychiatric diseases and in acute and chronic neurodegenerative disorders alike, based on the hypothesis that these conditions may represent a dysregulation of glutamatergic transmission, in particular the one mediated by NMDARs [36–38]. D-Serine has been associated with human disorders of the CNS both when they resulted in a decrease in neuromodulator level and neurotransmission, e.g., in schizophrenia, and in neuronal damage caused by increased levels, e.g., in ischemia and amyotrophic lateral sclerosis [39, 40].

Notably, a link has been identified between glycolysis and the phosphorylated pathway. In astrocytes, D-serine production is modulated by the interaction between SR and glyceraldehyde 3-phosphate dehydrogenase, the glycolytic enzyme producing 3-phosphoglycerate, which is the starting compound for L-serine synthesis [41]. The glycolytic flux controls D-serine synthesis: NADH inactivates SR (by allosterically affecting its interaction with ATP), and ATP increases the interaction between SR and glyceraldehyde 3-phosphate dehydrogenase, inactivating the racemase and decreasing D-serine production.

L-serine represents the precursor of phosphoglycerides, glycerides, sphingolipids, and phosphatidylserine. The external supply of L-serine is essential for the synthesis of sphingolipids and phosphatidylserine in cultured neuronal cells [42]. These phospholipids constitute important lipid messenger molecules in apoptosis signaling pathways, and gangliosides derived from sphingosine are relevant membrane and myelin components and are involved in cellular differentiation, proliferation, and migration.

The phosphorylated pathway is also strictly linked to cancer, since the hyperactivation of the serine/glycine biosynthetic pathway drives oncogenesis; for a review, see [43]. It has been proposed that increased L-serine levels may promote altered cellular proliferation, providing the building blocks and carbon units for the synthesis of nucleotides, lipids, and other cellular components. Cancer cells convert ≈ 10% of the 3-phosphoglycerate generated from glycolysis into the serine precursor 3-phosphohydroxypyruvate [44]: PSAT uses this molecule to convert glutamate to α-ketoglutarate that refuels the Krebs cycle, representing ≈ 50% of the anaplerotic flux and sustaining cancer cell metabolism. The glycine cleavage system refuels one-carbon metabolism, a complex, cyclic metabolic network based on chemical reactions of folate compounds. The upregulation of PHGDH and the synthesis of L-serine have been considered necessary and/or sufficient to promote both cancer cell growth and oncogenic transformation [43, 45]. In several types of leukemia and lymphoma, the increased PHGDH activity was due to an upregulation at the transcriptional level [46], while in breast cancer and melanoma, this could be due to a genomic alteration that provides a copy-number gain of the PHGDH gene [45]. In non-small-cell lung cancer cells, PSAT also appeared to be upregulated: it was able to modulate the activity of the transcription factor E2F, controlling the expression of genes essential for cell cycle regulation [47]. The PSAT-encoding transcript was overexpressed in colon adenocarcinoma [48] and increased with tumor progression in colorectal cancer [49]; high levels of the transcript in breast cancer were associated with a poor clinical response to endocrine therapy [50]. This topic is beyond the scope of this review, but we would like to highlight that serine metabolism provides opportunities for novel drug development [51] and biomarker identification [52] in cancers, too.

Phosphoglycerate dehydrogenase

Structure–function relationships

D-3-phosphoglycerate dehydrogenase is a cytosolic enzyme that catalyzes the reversible NAD⁺-coupled oxidation of the glycolytic intermediate 3-phospho-d-glycerate to 3-phosphohydroxypyruvate, the first step in the phosphorylated pathway, Fig. 2a [53]. The equilibrium of the reaction is almost in the direction of D-3-phosphoglycerate: less than 5% of the substrate/product is in the form of 3-phosphohydroxypyruvate. The reaction is driven toward 3-phosphohydroxypyruvate due to its consumption by downstream pathway steps. This provides a mechanism that keeps 3-phosphoglycerate from following the steps in glycolysis using it in the phosphorylated pathway only when the synthesis of serine is required [54]. PHGDH is a member of a family of proteins classified as 2-hydroxyacid dehydrogenases that are generally specific for substrates with a d-configuration [55].

Fig. 2.

Reactions of the phosphorylated pathway. a Phosphoglycerate dehydrogenase (PHGDH) catalyzes the NAD⁺-dependent conversion of d-3-phosphoglycerate into 3-phosphohydroxypyruvate; b phosphoserine aminotransferase (PSAT) catalyzes the transamination of 3-phosphohydroxypyruvate to 3-phosphoserine with glutamate as amino donor using a pyridoxal-5′-phosphate (PLP) cofactor; c phosphoserine phosphatase (PSP) catalyzes the irreversible hydrolysis of 3-phosphoserine to L-serine

This enzyme has been widely investigated in a variety of organisms, especially from the prokaryotes Escherichia coli and Mycobacterium tuberculosis; for a review, see [53, 54]. Eukaryotic PHGDHs have been described from Saccharomyces spp.[56], Entamoeba histolytica [57, 58], and from some higher plants. A human PHGDH-encoding transcript has been detected at high levels in prostate, testis, ovary, brain, liver, kidney, and pancreas, whereas lower levels are present in thymus, colon, and heart [59]. The human enzyme (hPHGDH) is involved in serious/lethal diseases (see below): despite its physiological importance, no detailed biochemical studies of hPHGDH have been reported so far.

The nucleotide sequence of human PHGDH gene was determined by [60]: it encodes a 533 amino acid protein (56.8 kDa), and shares a 94.6% and 36.8% sequence identity with PHGDH from rat and M. tuberculosis, respectively. Three types of PHGDH have been reported, referred to as types I, II, and III, differing in size and domain composition. hPHGDH belongs to type I, as do the homologues from M. tuberculosis, chicken, mouse, rat, and rabbit. All three types of PHGDHs contain two common domains: the substrate-binding domain and the cofactor-binding domain, located at the amino terminal part and involved in substrate binding and catalysis. Type I enzymes contain two additional regulatory domains, at the carboxy terminal end (Fig. 3a): the ACT (aspartate kinase–chorismate mutase–TyrA prephenate dehydrogenase) and ASB (allosteric substrate binding) domains. In some species, the ACT domain is a regulatory domain that acts as a binding site for L-serine to drive a negative feedback, although this regulatory mechanism could not be confirmed for hPHGDH. The ASB domain, conversely, impairs an additional level of allosteric control. hPHGDH is predicted to function as a tetramer, like the enzymes from M. tuberculosis and rat. The only available hPHGDH structure (PDB 2g76) is crystallized as a dimer, probably because a truncated species lacking the C-terminal domains involved in inter-subunit interfaces has been used [61]. In hPHGDH, the active site is covered by several loops from the first monomer (i.e., R54-V59, A76-V83, N97-G101, G152-L153, D175-I178, H206-L216, C234-V240, D260-D269, and C281-S287) and one loop from the second monomer (i.e., W133′-K136′). The side chains of the basic residues are predominantly oriented toward the active site for binding to the negatively charged substrate. Specifically, R53 interacts with the carboxyl group of the substrate, while R134 and R235 bind the phosphate group and the substrate is oxidized by the transfer of a hydride ion to NAD⁺ and a proton to H282.

Fig. 3.

Structure of the enzymes of the phosphorylated pathway with the residues linked to pathological disorders shown in red. a Left: structure of human PHGDH (pdb 5n6c). The substrate-binding domain is shown in blue, the NAD-binding domain in green. Right: details of the active site of hPHGDH in complex with NAD⁺ (orange) and l-tartrate (yellow); b Left: homodimer structure of hPSAT (pdb 3e77). The large domain is shown in blue, the small one in green. Right: details of the active site of hPSAT in complex with PLP (orange) covalently bound to K200; c Left: structure of the monomer of hPSP in the absence of an active site ligand (pdb 1nnl). The oligomerization domain is shown in blue. Right: detail of the Ca²⁺ ion coordinated in the active by D20, D22, and D179 and three water molecules; dashed lines represent H-bond metal–ligand interactions

The first characterization of hPHGDH demonstrated that it possesses a broad substrate acceptance, similar to that reported for the enzyme from E. coli but different from the rat homologue: it uses 2-oxoglutarate to produce the oncometabolite 2-hydroxyglutarate and to convert oxaloacetate to malate, Table 1 [62]. Several years later, the binding of d-3-phosphoglycerate and other analogs similar in size and containing the 2-hydroxypropanoic acid moiety (such as d,l-malate, d,l-lactate, l( +)tartrate, and α-ketoglutarate) to hPHGDH was evaluated using differential scanning fluorimetry, demonstrating that 3-phosphoglycerate (alone) and d,l-malate (in the presence of NAD⁺) stabilize the enzyme [61]. Recently, owing to the crucial involvement in tumors, hPHGDH inhibitors were identified. First attempts were undertaken in 2015, when AstraZeneca followed a fragment-based lead-generation approach starting from under-represented chemotypes and ring systems to derive a series of structurally simple fragments from indole. The result was composed of multiple fragments that bind in a competitive mode to the adenine region of the NAD-binding site: the compound FL2 exhibiting a K_d of 0.18 μM was identified, Table 1 [63]. Two noncompetitive inhibitors were identified by a high-throughput screening approach, starting from actual drugs following smart fragmentation [64, 65]: several selected compounds contained a nitrogen moiety such as morpholine, piperazine, or cyclohexamine, which is often attached to a thiocarbonyl. Based on six selected hPHGDH inhibitors containing a piperazine-1-thioamide scaffold, one improved compound was identified, which was called NCT-503 and showed an IC₅₀ of 2.5 µM, Table 1 [64]. Similarly, starting from a library of drug-like compounds, a total of 408 inhibitors were identified using a coupled in vitro enzymatic assay: seven compounds were selective for hPHGDH versus other NAD⁺-dependent dehydrogenases (the CBR-5884 compound showed an IC₅₀ = 33 μM) [65]. Later, other groups used the same approach of screening hits libraries followed by medicinal chemistry [66–70]. Starting from the indole-2-carboxamides and following several steps of optimization that took into account the shape of the adenosine- and phosphate-binding regions, the best allosteric, NAD⁺-competitive inhibitor discovered so far was identified: BI-4924 showed an IC₅₀ of 2 nM, see Table 1 [71]. A different approach was used by Unterlass and collaborators, who explored NAD⁺ analogs to discover novel inhibitors such as ADP-ribose that affected cofactor binding [61]. Recently, libraries of natural products were also screened to find molecules with anticancer activity acting as hPHGDH inhibitors, Table 1 [72, 73].

Table 1.

Kinetic properties and inhibitors of hPHGDH

The human PHGDH coding gene consists of 12 exons and has been mapped to chromosome 1p12. An analysis performed at the NCBI site with AceView highlighted that it contains 30 distinct GT-AG introns, with 8 probable alternative promoters, 5 nonoverlapping, alternative, last exons, and 8 validated, alternative, polyadenylation sites. According to AceView, transcription could produce up to 19 different mRNAs, among them 17 alternatively spliced variants and 2 unspliced forms. The length of alternative isoforms varies from 533 to 53 amino acids. The isoforms a (corresponding to the wild-type protein), b, g, and i are not truncated at the N-terminal. Among the others, two putative splicing variants appear to be particularly relevant, since they have been detected in brain or neuroblastoma cells: the first one (PHGDH.g) is 410 amino acids in length and lacks the sequence encoded by exons 10 and 11 in the C-terminal region; the second one (PHGDH.i) consists of 5 exons encoding for a 213 amino acid protein variant which contains an alternative splice variant of exon 5 and completely lacks the C-terminal regulatory domain. Both of these isoforms were confirmed by interrogating the AspicDB site. Interestingly, a shortened form of rat PHGDH lacking the C-terminal 209 amino acids has already been expressed [74]: the observation that this 36-kDa, truncated form is an active dimer points to a role of the C-terminal domain in tetramerization of the enzyme.

Phosphoserine aminotransferase

Structure–function relationships

The second enzyme in the L-serine synthetic pathway, phosphoserine aminotransferase, catalyzes the reversible glutamate-linked transamination of 3-phosphohydroxypyruvate to 3-phosphoserine, concomitantly producing 2-oxoglutarate, Fig. 2b. It is part of the large α-family of PLP-dependent enzymes and belongs to the group IV aminotransferases. The crystal structure at a resolution of 2.5 Å of the human enzyme (hPSAT) complexed with PLP (PDB 3e77, residues L17–L370) is characterized by an α/β structure and consists of a homodimer in which each monomer contains one PLP molecule and contributes to the active sites at their interface (Fig. 3b, left). The subunit’s interface is extensive: upon dimerization, approximately 12% of protomer surface is buried. Two domains can be identified in the individual monomers: the large domain (residues from L17 to G263) consists of six α-helices and a seven-stranded β-sheet, in which all strands except the second one are parallel (as observed in all members of α-family); the small domain consists of a two-stranded antiparallel β-sheet and three α-helices.

A superimposition of the structure of PSATs from different organisms shows that their overall organization is very similar, but the last two α-helices in the small domain vary in length and, in hPSAT, the loop between the first two α-helices in the same domain is longer [75]. This loop is close to the one between the last two α-helices, and both loops are located at the entrance to the active site cleft: the longer loop in the human enzyme is thought to endow a greater flexibility, resulting in better access to the active site and thus affecting the biochemical properties [75]. Analogously to the enzyme from E. coli (EcPSAT) [76], the two active sites in the dimer are formed by residues belonging to both the large and the small domain, implying that the dimeric configuration is essential for functional activity. The PLP cofactor is covalently bound to K200, forming an internal aldimine moiety, with its pyridine N atom placed towards the large domain and hydrogen bonded to D176, W107, and T156 (Fig. 3b, right). These interactions contribute to fixing the aromatic ring of PLP parallel to the indole ring of W107, which provides a favorable stacking interaction. Other highly conserved residues, namely, S43, H44, R45, and R335, are present at the hPSAT active site and are involved in substrate binding, as indicated by the structure of Bacillus alcalophilus PSAT complexed with l-phosphoserine [77]. Characterization of deletion variants of PSAT from the protozoan parasite Entamoeba histolytica revealed a relevant role of the N-terminal residues. A short N-terminal loop (corresponding to the M1-K16 portion in hPSAT) is largely buried in the active site of the wild-type enzyme where it acts by restricting the movement of the loop containing the lysine residue that forms the internal aldimine with the PLP cofactor and by providing the architecture to the active site for binding the incoming substrate. The deletion of even some of the residues within this loop results in the almost complete loss of enzyme activity [75].

Little is known about the substrate preference of hPSAT. Bovine PSAT can use 2-amino-4-phosphobutyrate, 2-amino-5-valerate, and homocysteate as substrates [78], and EcPSAT accepts l-aspartate [79]. The structure of EcPSAT complexed with α-methyl-l-glutamate suggested that l-aspartate binding might be stabilized by several arginine residues in the active site that can provide ion pairs and H bonds with this small molecule [76]. Analogously to other PSATs, the human enzyme shows a localized, positive charge around the active site, while a negative charge is dispersed all over the surface. The hPSAT kinetic process likely proceeds through a ping–pong bi–bi mechanism, as reported for the enzymes from cow and Arabidopsis thaliana [78, 80], and is affected by a substrate inhibition effect [81]. In this regard, recombinant bovine PSAT is subjected to product inhibition by l-phosphoserine [78]. Effective and specific hPSAT inhibitors are currently not available.

In humans, high levels of PSAT-encoding transcript have been detected in brain, liver, kidney, and pancreas, whereas lower levels were determined in thymus, prostate, testis, and colon [59]. Therefore, as compared to the hPHGDH-encoding gene (see above), the hPSAT gene exhibits a more restricted tissue-specific expression, although both genes are abundantly expressed in the brain. Worthy of note is that the primary transcript encoding for hPSAT undergoes an alternative splicing process, producing two different mature isoforms, namely, hPSATα and hPSATβ proteins. The difference between hPSATα (324 amino acids, 35.2 kDa) and hPSATβ (370 amino acids, 40.0 kDa) is the presence of 46 residues in the V290-S337 region, encoded by exon 8, which is lacking in the shorter protein isoform. hPSATα is not present in other organisms. Additionally, among several analyzed human cell lines, the hPSATα transcript was detected only in K562 and HepG2 cells, which appeared to barely express the protein isoform, while it was never expressed at detectable levels in healthy tissues [59]. On the other hand, hPSATβ was highly expressed in all the analyzed cell lines and in several human tissues, indicating that this protein isoform is involved in the L-serine biosynthetic pathway, at least under physiological conditions. This statement was further supported by the evidence that the recombinant hPSATα protein is poorly active [59].

According to AceView, the hPSAT-encoding gene contains 12 distinct gt-ag introns; 2 nonoverlapping, alternative, last exons, and 6 validated, alternative, polyadenylation sites are found. Transcription is predicted to produce seven mRNAs (five alternatively spliced variants and two unspliced forms). The isoform bAug10 corresponds to hPSATα. A transcript might generate a longer isoform of 457 amino acids (PSATlong, 50 kDa): it should contain an additional translated region at the N-terminal of the protein sequence arising from an alternative starting codon. Interestingly, an analogous isoform has been predicted in monkeys (accession code XP_002819935).

Phosphoserine phosphatase

Structure–function relationships

Phosphoserine phosphatase catalyzes the third, irreversible, and Mg²⁺-dependent step of L-serine biosynthesis (Fig. 2c). PSP is a member of the haloacid dehalogenase superfamily, forms an acylphosphate during catalysis, and shares three motifs with P-type ATPases and haloacid dehalogenases, which are located at the active site [82]. The residue phosphorylated during catalysis in the human enzyme (hPSP) was identified as D20, i.e., the first aspartate in the first conserved motif (DXDXT) [83]. The second motif contains a conserved serine/threonine residue and the third motif is constituted by a conserved lysine, followed by less conserved residues, and a conserved aspartate. Mutagenesis studies showed that all three motifs play an important role in catalysis [83]. Substitutions S109T, G178A, D179E, and D183E affected hPSP activity only to a minor extent, while the S109A, D179N, G180A, D183N, and K158A/R substitutions resulted in > 80% decrease in activity. The G178A and D179N variants showed decreased affinity for the substrate phosphoserine and the D179E/N decreased affinity for Mg²⁺ [83]. Compared to the wild-type enzyme, the relative activity was decreased by ≥ 80% for S23T, E29D/Q, R65K/A, N133A, T182S, and R202K/A hPSP variants [84]. Recently, it was suggested that when D183 is substituted the enzyme is rearranged and Thr182, a residue directly bound to every ligand in hPSP, is retracted and this affects the substrate stabilization [85].

At first, the structure of hPSP was solved at a resolution of 1.5–2.8 Å in complex with the product L-serine and the competitive inhibitor 2-amino-3-phophonopropionic acid [84], as well as in the absence of a ligand [86]; see Fig. 3c. hPSP consists of two main domains: the first one resembles a Rossmann fold and the second consists of the monomer–monomer interface (β4 and β5) and a helix bundle (α2, α4, and α5). A high degree of flexibility is apparent considering the orientation of a helix in the helix bundle portion: the hPSP dimeric enzyme in complex with an inhibitor showed a protomer in the closed conformation and the second one in an open conformation (the semi-open state, in which the helix covering the active site is visible), while in the absence of an active site ligand, an even more open conformation was apparent. In the latter state, the enzyme is partially unfolded to allow the substrate access to the active site: substrate binding causes hPSP to shift to the closed conformation by stabilizing the partially unfolded region. Three new crystal structures of hPSP in complex with the substrate phosphoserine, the product L-serine, and homocysteic acid showed that such a loop can be also present in a totally unfolded conformation [85].

The reaction catalyzed by hPSP is Mg²⁺ dependent, although the affinity for the metal cofactor is unclear [82, 87]. Other di- or tri-valent ions inhibit enzymatic activity (e.g., Ca²⁺, Mn²⁺, and Al³⁺) even in the presence of magnesium ions. The Ca²⁺ ion bound to the active site of hPSP is coordinated by three water molecules and three aspartate residues (D20, D22, and D179); see Fig. 3c, right. When Ca²⁺ replaces the Mg²⁺ ion at the active site, the metal–ligand-binding pattern is altered: Ca²⁺ captures both side-chain oxygen atoms of D20, while a Mg²⁺ ion ligates only one oxygen atom of this residue. This alters the orientation of l-phosphoserine required for catalysis and hampers the nucleophilic attack of one of the D20 side-chain oxygen atoms on the phosphorus atom of the substrate [82].

The reaction of hPSP starts with phosphoserine binding: its NH₂ group interacts with the side chain of E29 and an H bond is formed between the substrate-COOH group and the side chain of R65. Based on the newly solved 3D structures and the distance between the substrate and R65, the role of the latter residue was recently investigated [85]. Accordingly, D20 carries out a nucleophilic attack on the scissile phosphate: phosphoserine is then cleaved, resulting in the release of serine (D22 serves as a general acid donating a proton to serine) and the formation of a covalent phosphoaspartyl intermediate with D20. A water molecule takes that position in the just-vacated group site and D22 extracts a H⁺ from this water molecule and can then carry out a nucleophilic attack on the phosphoaspartyl intermediate. The catalytic cycle is completed when the inorganic phosphate is dissociated by opening the enzyme. As stated above, the Mg²⁺ ion is essential for phosphoserine hydrolysis, playing both a catalytic role in the reaction by orienting/stabilizing D20 in an optimal position to attack the phosphorus atom of phosphoserine and by its positive charge, which facilitates the nucleophilic attack of D20 by extracting the negative charge from the phosphate group [82].

hPSP purified from brain dephosphorylated both l- and d-phosphoserine, with a higher affinity for the l-enantiomer [88]. hPSP was inhibited by its product L-serine, suggesting that the whole synthetic pathway might be regulated at this level by the demand of the end product. D-serine had no inhibitory effect, implying that d-phosphoserine is unlikely to be the physiological substrate, also given that it is reportedly absent in rat brain [89]. hPSP showed significant activity on other phosphoesters, such as fructose-6-phosphate, β-glycerophosphate, and phospho-l-tyrosine. Recombinant hPSP was expressed in E. coli and purified as a 50-kDa homodimer: specific activity was 24.5 U/mg protein [90]. Unlike the results reported for the enzyme purified from human brain, recombinant hPSP did not dephosphorylate other phosphate esters [87].

A preliminary investigation reported that PSP activity is inhibited by p-chloromercuriphenylsulfonic acid and glycerophosphorylcholine (IC₅₀ = 9 and 18 µM) [91] and that rat PSP is inhibited by both the metabotropic glutamate receptor antagonist l-2-amino-3-phosphonopropionic acid (l-AP3) and its d-enantiomer (151 and 48 µM, respectively) [92]. hPSP isolated from brain was partially inhibited (20–40%) by 1 mM of antipsychotic drugs (such as trifluoperazine and chlorpromazine) and activated by benzodiazepines, such as diazepam and chlordiazepoxide (13–59%) [88]. A high-throughput screening identified ten compounds that inhibited M. tuberculosis PSP activity, but were not cytotoxic for mammalian cells: CID 7 (NSC 93739, Rosaniline) and CID 10 (NSC 165701) also inhibited hPSP activity by 70% (at 100 µM concentration) [93].

Distribution of PSP activity in brain differs between species [88]: in human brain, the highest activity was apparent in the hippocampus and the lowest in the cortex; in rat, the activity was roughly similar in all areas; and in rabbit, the cerebellum showed significantly lower PSP activity than the other areas.

According to the AceView site, the human PSP gene contains two probable, alternative promoters, nine nonoverlapping, alternative, last exons, and five validated, alternative, polyadenylation sites. Notably, 27 mRNAs might be transcribed (25 alternatively spliced variants and two unspliced forms), differing in: (1) the 5′ or 3′ end; (2) the presence or absence of 21 cassette exons; (3) overlapping exons with different boundaries; and (4) splicing versus retention of three introns.

Human defects in serine synthesis

Serine deficiency disorders were first described in 1996 as severe, infantile, neurological disorders, including congenital microcephaly, intractable seizures, and several psychomotor defects. These patients showed low concentrations of L-serine and glycine in plasma and cerebrospinal fluid, decreased white matter volume, and hypomyelination [94]. Subsequently, a milder phenotype was also found in childhood, which presents with developmental delay and intellectual disability [8], and in adulthood, with psychomotor retardation, congenital cataracts, slight cerebellar ataxia, and axonal polyneuropathy [95].

Neu–Laxova syndrome (NLS) was identified as a rare, severe, and lethal L-serine deficiency disorder originating from defects in all three genes of the enzymes of the phosphorylated pathway [13]. NLS manifests with severe malformations, leading to prenatal or early post-natal death, and is genetically heterogeneous [96]. Children present with dysmorphic features (proptosis of the eyes, abnormal eyelids, small round mouth, microcephaly, and extensive skeletal abnormalities with contractures and webbing of fingers and toes), skin abnormalities, and structural abnormalities of the CNS (with neural tube defects, cortical dysplasia, enlarged ventricular spaces, and alterations of the cerebellum). This pathology is classified into three phenotypes based on the onset of the disease [97]: (1) infantile, affecting the majority of children with serine deficiency: it is mainly due to phosphoglycerate dehydrogenase deficiency; (2) juvenile, at present affecting only one family with a milder serine deficiency phenotype [8]; and (3) adult: a single adult patient (with hPHGDH deficiency) was diagnosed with mild mental disability and mild cerebellar ataxia in childhood, but developed progressive polyneuropathy in adulthood [95].

PHGDH mutations and pathological states

Altered expression and/or activity of hPHGDH have been correlated to several pathological conditions [53, 98, 99]: serine deficiency disorder [8, 94, 95, 100–102], NLS [13, 96, 103–106], and cancer [43, 107–109]. Altered hPHGDH activity resulting in a decrease in L-serine levels in CSF and plasma has been correlated to neurological disorders: several genetic mutations have been reported and the enzyme deficiency confirmed in cultured skin fibroblast from patients [13].

In 2000, two hPHGDH variants (i.e., V425M and V490M) were identified in six patients from four different families [100], previously reported in [84, 110, 111]. All the patients presented with congenital microcephaly, psychomotor retardation, intractable seizures, and decreased L-serine concentrations in CSF and plasma. Assays performed on the in vitro translated protein variants or using patient-derived fibroblast lysates concluded that V425M and V490M substitutions did not affect the protein synthesis or the turnover of hPHGDH, but resulted in an ≈ 50% decrease in maximal activity with no change in K_m values [100]. A contrasting conclusion arose from studying siblings displaying a 70% reduction in enzyme activity and harboring the V490M substitution [101]: using CHO and BHK cells and the recombinant proteins expressed in E. coli, the authors concluded that this substitution reduced the amount of active enzyme present in solution, due to an impairment in the protein folding pushing its degradation [101]. The V425M substitution, together with the R491W substitution, was then detected in a 31-year-old patient displaying a milder phenotype [95] and the V490M in one case of NLS [103]: no biochemical analyses were performed in these patients. These substitutions are located in the regulatory domains of the hPHGDH. V425 is located in the ASB domain, in the middle of a β-sheet: its substitution could affect the local secondary structure. V490M and R491W are in the ACT domain at the dimer interface and could affect dimerization.

Since 2009, several mutations have been correlated to hPHGDH deficiency [8, 99, 102, 112, 113]. Five homozygous missense mutations have been identified in eight different individuals with congenital microcephaly, psychomotor retardation, and variable presence of seizures (encoding R135W, V261M, A373T, G377S, and G429V variants) [8, 99, 102, 113]. The R135W substitution was also found in an additional patient displaying the same phenotype: in this case, a frameshift mutation (p.G238fsX) coding for a premature stop codon and the loss of the last 295 residues was also detected [102]. Activity assays were performed on patients’ skin-derived fibroblasts and on HEK293T cells expressing hPHGDH variants: all the missense mutations resulted in decreased activity due to an increased K_m value in the case of V261M variant (fourfold higher than the wild-type PHGDH) and a decrease in V_max for the others [102]. The R135W substitution was also detected together with the R163W substitution in a 4-year-old patient with microcephaly, intrauterine growth retardation, severe psychomotor delay, and intractable seizures [112]. The patient displayed very low concentrations of serine in CSF, hypomyelination, and corpus callosum atrophy. After treatment with L-serine up to (650 mg/kg/day) for 1 week, the seizures completely ceased.

The G377S substitution (inherited in a heterozygous way) together with a splice-site mutation (c.138þ2dupT) was also identified in two 8-year-old sisters with microcephaly and developmental delay [113], displaying low L-serine and glycine levels in CSF and mildly (16%) reduced hPHGDH activity. Oral treatment with serine (500 mg/kg/day) and glycine (250 mg/kg/day) restored the amino acid concentrations in CSF and ameliorated the spasticity and responsiveness, but had no effect on developmental progress [113].

From a structural point of view, the residues R135, R163, and V261 are located in the nucleotide-binding domain. R135 interacts with the cofactor: substituting it with a tryptophan could disrupt this interaction. R163 is located at the dimer interface, while V261 is located in a loop close to the cofactor and its substitution could result in a steric clash with surrounding residues. A373, G429, and G377 are located in the ASB domain.

In the last decade, several novel, missense (R54C, S55F, G140R, R163Q, L249P, E265K, A286P, C421W, G429V, and R491W), nonsense (c.1518G > A, p.Trp506*), and frameshift (p.Gln433* and Met477Asnfs*51 encoding for a premature stop codon at position Q433 and M477, and the c.1A > C mutation which abolishes the start codon) mutations of the human PHGDH gene have been identified and correlated to NLS; see Table 2 [13, 96, 103, 104, 114–116]. However, no biochemical studies have been performed yet.

Table 2.

Uncharacterized hPHGDH variants related to NLS and their localization in protein structure

hPHGDH variants	Localization	References
R54C	In the substrate-binding site	104
S55F	In the substrate-binding site	115
G140R	At dimer interface, in the nucleotide-binding site	13, 96
R163Q	At dimer interface, in the nucleotide-binding site	107
L249P	In the nucleotide-binding site	114
E265K	In the nucleotide-binding site	104
A286P	In the substrate-binding site	104
C421W	In the middle of a β-sheet in the ASB domain	103
G429V	In the middle of a loop in the ASB domain	13

A recent investigation reported a decrease in hPHGDH levels in human hippocampal astrocytes during the progression of Alzheimer’s disease (AD): astrocytic L-serine biosynthesis was impaired in young AD mice and in AD patients [117]. The lower production of D-serine could explain the observation that AD mice displayed a lower occupancy of the NMDAR coagonist site as well as synaptic and behavioral deficits. The authors suggested an altered glycolysis flux in early AD as the primary event leading to less serine production and a lower hPHGDH expression in later AD stages. Supplementation with L-serine in the diet prevented both synaptic and behavioral deficits in the mice model of AD.

PSAT mutations and pathological states

The first two cases of hPSAT deficiency were reported in 2007 [118]. The disorder was identified in two young siblings who showed low concentrations of serine and glycine in plasma and cerebrospinal fluid (CSF) and was clinically characterized by intractable seizures, acquired microcephaly, hypertonia, and psychomotor retardation. Prognosis was poor once the patient developed symptoms, but treatment with serine and glycine supplementation from birth led to normal development [118]. PSAT activity in the two patients was lower (≈ 50%) than in healthy controls. Genome sequencing revealed that both affected siblings were heterozygotes for mutations in the hPSAT-encoding gene: a frameshift mutation in exon 2 (c.delG107) and a missense mutation (c.299A > C) yielding the D100A substitution were identified [118]. The recombinant D100A hPSAT showed a low solubility (the yield in soluble protein was ≈ tenfold lower than in the wild-type enzyme) and low enzymatic activity (although the K_m for the substrate remained unchanged, the V_max was ≈ sixfold decreased) [118]. As the highly conserved D100 is thought to form H bonds with the surrounding residues to hold a neighboring loop in place, the authors inferred that the observed effects were caused by the disruption of this hydrogen-bonding network [118].

Mutational analysis performed in Northern Europe identified six individuals affected by NLS who showed mutations in PSAT1 gene [104]. Two types of mutations were detected: a frameshift mutation (a homozygous indel—c.1023_1027delinsAGACCT) and different missense mutations. The recurrent homozygous (c.296C > T) and heterozygous (c.536C > T) change yielded A99V and S179L substitutions, respectively, as well as the concomitant presence of the two substitutions. Based on molecular modeling analyses, the authors predicted that the substitutions deeply affected enzyme function by altering the PLP or substrate binding, thus hindering the transamination reaction [104].

Alterations in hPSAT expression and activity were also related to psychiatric disorders. In 2011, the case of two related individuals (mother and son) suffering from schizophrenic spectrum condition and carrying a hereditary, balanced chromosomal translocation was studied and the PSAT1 gene was identified as a susceptibility gene [119]. In both patients, a significantly reduced expression of hPSAT was associated with a marked decrease in serum levels of L-serine and glutamate compared to healthy controls.

PSP mutations and pathological states

hPSP defects have been linked to the onset of neurological disorders, and a number of SNPs that alter the enzymatic activity have been identified.

In a patient with Williams’ syndrome (characterized by pre- and post-natal growth retardation, moderate psychomotor retardation, and facial dysmorphism), two mutations in human PSP gene were identified in which a well-conserved aspartate was replaced by an asparagine (D32N) and an extremely conserved methionine by a threonine (M52T) [120]. Recombinant proteins expressed in E. coli showed, respectively, a specific activity of 11.5 and 0.8 U/mg vs. 24.5 U/mg for the wild-type hPSP [90]. Interestingly, the patient showed decreased CSF serine levels, while phosphoserine and glycine levels were normal [90]: oral serine administration normalized serine levels in plasma and CSF and seemed to have some clinical effect. A benefit from oral serine supplementation (2.5 g orally, three times a day) was also observed in a female, adult patient characterized by the V44G and G141S hPSP substitutions and showing intrauterine growth restriction, lifelong intellectual disability, childhood-onset epilepsy, and borderline microcephaly [121]. After 4 months of treatment, serine and glycine plasma levels reached normal values and the patient’s condition improved significantly. In a family in which individuals had moderate to profound intellectual disability and seizures, the A35T hPSP substitution was identified. The specific activity of this variant (2.5 U/mg) was ≈ tenfold lower than that of the wild-type enzyme [122]. A homozygous frameshift mutation (c.267delC, p.Gly90Alafs*2) in the gene encoding hPSP was also reported in a family in which a diagnosis of NLS was ascertained based on clinical features [104].

An alteration in L-serine metabolism was also proposed in patients with schizophrenia owing to increased hPSP activity as compared to controls, especially in male subjects, although mRNA levels were lower [123]. Furthermore, hPSP activity negatively correlated with plasma D-serine and glycine concentrations and D-serine/total serine ratio, which positively, albeit weakly, correlated with plasma L-serine concentration, especially in male patients. A link between PSP and neuronal function in AD has also been proposed [124]. Actually, β-amyloid activated PSP in rat neuroblastoma B104 cells and thus enhanced glutamate-induced neurotoxicity by acting on D-serine level: the PSP increase was inhibited by interleukin-11, which exerted a protective effect against β-amyloid-induced neurotoxicity in a dose-dependent manner.

Conclusions

Defects in serine biosynthesis in humans highlight the role of de novo production of this nonessential amino acid in brain development and morphogenesis. Patients with a serine deficiency show severe neurological symptoms. Furthermore, we have learned in recent years that perturbations in the glutamatergic tripartite synapse may underlie the pathogenic mechanism of AD [125] and of psychiatric diseases such as schizophrenia [36]. It is, therefore, extremely important to define the molecular mechanisms involved in modulating brain concentrations of the enantiomers of serine and the interplay between glial and neuronal cells. We reported here that, although the three mammalian enzymes of the phosphorylated pathway have been heterologously expressed and their structures have been solved, the mechanisms regulating the enzymes’ activity at the physiological level are largely unknown. Actually, although the pivotal role of PHGDH in L-serine (and indirectly D-serine) synthesis has been recently demonstrated through the generation of knock-out mice [12], the effects of modulating the other enzymes involved in the L-serine biosynthetic pathway have not been elucidated yet.

As Tom de Koning recently stated about serine deficiency disorders, “it is not possible to discriminate the different gene defects on clinical grounds. Molecular defects in the genes encoding the three enzymes can present with identical phenotypes ranging from a severe lethal antenatal phenotype to a milder adult onset polyneuropathy phenotype. However, recognition of serine deficiency is important, because good treatment results have been reported with L-serine therapy” [97]. Accordingly, an in-depth investigation of the three enzymes of the phosphorylated pathway is mandatory to identify the most suitable therapeutic approach. Treatments should also focus on the potential benefits of supplementing serine (or serine analogs). The recent proposal to use oral L-serine supplementation as a ready-to-use therapy for AD [117] is opening a new, promising avenue for a number of pathological conditions.

Taken together, elucidating the mechanisms affecting the function of enzymes belonging to the phosphorylated pathway and those related to serine metabolism (namely, SR, DAAO, and SHMT) might provide key information for designing specific molecular approaches/tools by which cellular concentrations of serine could be finely tuned.

Code availability

Not applicable.

Author contributions

All authors designed, wrote, and revised the review.

Funding

This work was supported by a grant from Ministero Università e Ricerca Scientifica PRIN 2017 (Grant 2017H4J3AS) to LP, and from Fondo di Ateneo per la Ricerca to LP and SS.

Data availability

Not applicable.

Compliance with ethical standards

Conflict of interest

The author(s) declare that they have no conflicts of interest.

Ethics approval

Not applicable.

Consent to participate

Yes.

Consent for publication

Yes.