Modelling helps to identify a rare childhood neurological disorder caused by missense mutations of SynGAP1 protein

The figure panels show, from left to right, how a pathogenic missense mutation alters the structure of the SynGAP1 proteinA research project led by Pekka Postila models the effects of missense mutations on the structure and function of the SynGAP1 protein. Malfunction of this protein causes a variety of symptoms such as severe intellectual disability, epilepsy, and autism. The research work demands a meticulous approach, a great deal of patience and enormous computational power.

Sometimes, a change in just a single nucleotide in the genome can lead to significant health problems and the onset of disease. When such a change swaps one amino acid for another in a protein encoded by an important gene, it is referred to as a missense mutation. Such a change may look insignificantly small, but in reality it can be devastating to the protein’s vital function.

One gene that is particularly sensitive to such changes is SYNGAP1, which encodes SynGAP1 protein that is abundantly expressed in neurons of the forebrain. The SynGAP1 protein participates in the regulation of intracellular growth and thereby inhibits, for example, tumorigenesis or cancer development. It is a large protein consisting of more than 1,300 amino acids, so the growth regulation is not its only or necessarily even its most important function in the human body.

“In recent years, the complex biology of SynGAP1 has finally begun to become clearer,” says Pekka Postila, Senior Research Fellow and Docent at the MedChem.fi laboratory, Institute of Biomedicine, University of Turku.

SynGAP1 acts as a kind of gatekeeper of the synapse. When we learn something new, SynGAP receives a chemical signal and steps aside for a moment. Then the relevant receptors can move into place, and communication between neurons is strengthened.

Rare neurological disorder in children

Neurological symptoms caused by SynGAP1 mutations are rare: the disease occurs in only about six per 100,000 people. The disease is detected in early childhood, and it affects the development of the nervous system decisively throughout life.

In our genome, each gene has two copies, one from the mother and one from the father. For this reason, an error in one gene copy does not always necessarily cause significant problems. In the case of the SYNGAP1 gene, however, the situation is different: reduced expression of just one gene allele is enough to cause a severe neurological disorder.

The clinical picture is diverse: typical symptoms include varying degrees of intellectual disability, epilepsy and features of the autism spectrum. Precisely because of this diversity, the disease is no longer referred to as SynGAP syndrome but simply SynGAP-related disorder.

Mutations in the SYNGAP1 gene are usually not inherited from the parents but arise in the child’s genome at the embryonic stage. There are several types of mutations, but they are generally divided into two groups: truncations and missense mutations.

Truncations shorten the SynGAP1 protein or can prevent it from being produced altogether. Such major changes are usually easy to identify with genetic tests. However, most of the harmful SYNGAP1 mutations are estimated to be small missense mutations, in which one amino acid is replaced by another.

A single amino acid substitution in a long amino acid chain does not always cause significant functional changes. The effect depends decisively on where in the protein’s three-dimensional structure the change occurs, and which amino acid replaces another. In one location the change may be harmless, but in another it may be disastrous for function.

For this reason, missense mutations are particularly challenging to diagnose. Moreover, there are currently no experimental methods by which the pathogenicity of SYNGAP1 missense mutations could be reliably verified. Precise computational methods are therefore needed to support diagnosis and assessment of the mutation effects on the protein’s function and structure.

Pekka Postila and his research group have received a total of 230,000 dollars in funding to study missense mutations from the California-based Cure SYNGAP1 foundation. The aim of the research is to improve the diagnostics of SYNGAP1 missense mutations by making particular use of computational structural research.

The mystery of folding

To understand missense mutations and their significance for protein structures, it is useful to look back in the history of protein structure determination briefly. All proteins are formed from an amino acid chain, whose links are amino acids connected to one another by peptide bonds. There are 20 different basic building blocks.

The building blocks of proteins and their significance for life were identified as early as the nineteenth century, but the precise structure of proteins was not understood before the 1950s. Then, researchers at the University of Cambridge, John Kendrew and Max Perutz, succeeded in depicting proteins for the first time three-dimensionally using X-ray crystallography.

The sight was a surprise: proteins resembled tangled piles of spaghetti or balls of yarn, full of many kinds of folds, braids and loops. The form, however, was not random. It is precisely the protein’s three-dimensional structure that determines its interactions with its environment: what the protein binds to, what it lets come close and what it repels. Kendrew and Perutz received the Nobel Prize for their pioneering work in 1962.

In 1961, the American researcher Christian Anfinsen showed that a protein’s shape is encoded directly in its amino acid sequence. An identical amino acid chain therefore always produces the same protein structure. This insight, too, led to a Nobel Prize, in 1972.

At the same time, a new mystery was born, which remained unsolved for decades: although protein amino acid sequences could be determined using spectrometry, no one understood by what mechanisms these chains fold into complex three-dimensional entities.

In 1994, a competition was launched with the aim of predicting the three-dimensional shapes of proteins from amino acid sequence alone. However, the first computer models were crude, and for more than two decades the accuracy of the protein structure predictions barely improved.

The situation changed in 2018, when neural network researcher Demis Hassabis and biochemist John Jumper introduced a new tool. AlphaFold, developed at Google’s DeepMind laboratory, uses machine-learning neural networks to search for similarities between the structures of known proteins and a new amino acid sequence.

AlphaFold predicts a protein’s shape by combining information from the solved structures of related proteins, much like traditional homology modelling. AlphaFold2 has since been used to predict the structures of more than 200 million proteins. Hassabis and Jumper were awarded the Nobel Prize in 2024.

SynGAP Missense Server: modelling missense mutations and assessing pathogenicity

The structure of the SynGAP1 protein produced by AlphaFold2 has been used to help model missense mutations. The method nevertheless has significant limitations regarding mutations.

“A learning neural network cannot anticipate mutations, and that’s where it can go well and truly wrong. Even if a mutation was known to distort severely the protein structure, AlphaFold still produces an intact and, so to speak, healthy model,” Postila explains.

AlphaFold alone therefore is not sufficient for assessing the effects of missense mutations. Acquiring a more accurate picture requires other computational methods and, hopefully, even new experimentally solved protein structures.

On the basis of the AlphaFold model, Postila’s research group has modelled all likely mutations arising from a single nucleotide substitution into the structure of the SynGAP1 protein. If reliable structural information was not available, the group used sequence-based and other non-structural prediction methods to assess pathogenicity.

The results produced by the modelling have been compiled in the SynGAP Missense (SGM) server, which provides an open online portal for assessing the pathogenicity of SYNGAP1 missense mutations.

At the structural level, the SGM server currently contains 3,325 missense variants and, at the sequence level, as many as 8,751. Based on the best methods, the database presents a consensus pathogenicity prediction, an AI-generated written prediction summary, links to clinical databases, and both a three-dimensional model and a two-dimensional interaction diagram for each missense mutation.

The SGM server and the database have been coded and are maintained by researcher Jukka Lehtonen from the Structural Bioinformatics Laboratory at Åbo Akademi University. Because the database was created at the request of a patient organization, it aims to meet a real diagnostic need.

Supercomputers help to uncover the truth about missense mutations

For SynGAP1 missense mutations listed in the clinical ClinVar database, Postila’s group has taken the structure-based analysis one step further. The mutations were not only modelled as static three-dimensional structures, but the structural changes caused by them were also investigated thoroughly using molecular dynamics (MD) simulations.

These atomistic simulations are based on a semi-empirical force field model, in which the atoms of amino acids are represented as if they were spheres and the bonds between these spheres act like small springs. When heat is added to the model, the atoms begin to move, the bonds flex and parts of the protein interact with one another. The protein begins to “live” or “breathe” in the same way as happens in real cells.

These computer simulations make it possible to determine precisely how missense mutations alter the protein’s three-dimensional structure and, importantly, predict which of them prompts big enough change to cause the neurological disorder.

Despite the apparent simplicity of the method, the computational load is enormous, and simulations even on the scale of nanoseconds require the use of supercomputers. In the research, 211 missense variants have been MD-simulated for a total of about 100 microseconds (150 ns × 3 × 211). In addition, folding stability calculations based on MD simulations have been performed for other variants present at the known structural parts as well. Based on benchmarking, the accuracy of the calculations is in a class of its own.

Without the high-performance computing provided by CSC – IT Center for Science, such extensive computation would not have been possible. Postila praises the collaboration with CSC profusely.

“At times when it feels as if everything is being cut, a centralized service organization like CSC is an excellent arrangement for a research group like ours. Hopefully this will not change in the future either.”

The project’s simulations have been run mainly on the GPU partition of the Puhti supercomputer. Postila’s group has, however, also used the Mahti supercomputer in simulations and AlphaFold modelling. Running one simulation covering about 140,000 atoms typically takes a couple of days, and in addition all simulations must be repeated at least three times.

“The model is not fully deterministic, and in successive simulations there can be small differences. Multiple runs ensure that we have understood correctly,” Postila notes.

Looking to the future: the interactions of the SynGAP1 protein

The research led by Postila actively aims to expand the SGM server to include the best information on assessing the pathogenicity of SYNGAP1 missense mutations. New prediction methods such as AlphaMissense, which is based on AlphaFold, are appearing at an accelerating pace.

At the structural level, the group is investigating how missense mutations affect the interactions of the SynGAP1 protein with other proteins or, for example, the cell membrane.

When examining gene sequences and protein structures as well as abstract pathogenicity predictions, it is easy to forget that proteins do not function alone. Inside the cell, they exist in constant, dynamic interaction with one another.

SynGAP1 is no exception in this regard: its coiled‑coil (CC) domain from each protein chain wraps around the neighboring chains, forming a three‑stranded bundle that stabilizes the trimeric assembly of SynGAP1.In addition to itself, SynGAP1 also binds to regulatory proteins such as Ras and Rap GTPases and to various scaffold proteins such as PSD95, which are central to neuronal signaling and synapse function.

“As computational power grows, we can increase our understanding by running ever longer simulations and modelling ever larger systems that include several interacting components,” Postila notes.

In the video (YouTube), Pekka Postila demonstrates how the SynGAP Missense Server works in practice.

Text, researchers’ group photograph and video: Juha Merimaa
27 February 2026
Read the article in PDF format