High dose L-Serine to treat children under 7 with severe autism + ID ? It works in Korea

 

Source: Joon Kyu Park, CC BY-SA 3.0 <https://creativecommons.org/licenses/by-sa/3.0>, via Wikimedia Commons

 

Today’s post is a follow up to the recent one that showed Memantine was beneficial to people with level 1 autism, normal IQ, with ADHD and anxiety/depression.

Our reader Hoang, highlighted a recent trial in Korea that used the OTC supplement L-serine, which has a biological effect that is the opposite of Memantine. The trial is part of series looking at treating those with severe autism with ID (intellectual disability). 

High-dose L-serine has been tested in children with severe autism and intellectual disability, and the main benefits were seen in those under 7 years old. While it may work by boosting NMDA receptor activity through conversion to D-serine, other brain-supporting roles of L-serine—like helping neuron membranes and reducing stress on brain cells—could also contribute. Older children may not respond as well, possibly because their brains are less plastic or they convert less L-serine to D-serine. Researchers should now explore whether direct D-serine dosing might help older kids, but safety must be considered.

 

The Trials and Target Group

The trials of AST-001, a syrup formulation of L-serine, focused on children with severe autism and intellectual disability (ID). The phase 2 study included children aged 2–11, but the most pronounced improvements were in those under 7 years old. The benefit did not entirely disappear after age 7, but it was smaller and harder to measure.

Dosing was weight-tiered:

Weight (kg)	Dose (g, twice a day)
10–13	2
14–20	4
21–34	6
35–49	10
>50	14

The outcomes measured were adaptive behavior (Vineland Adaptive Behavior Scales II) and clinical global impressions, with high-dose L-serine showing a statistically significant improvement over placebo.

 

How L-Serine Might Work

1. NMDA Receptor Modulation

L-serine can be converted in the brain to D-serine, a co-agonist of NMDA receptors, which are critical for learning, memory, and social behavior. This mechanism aligns with the idea that boosting NMDA signaling could help in some autism. This is the exact opposite of what Memantine does.

2. Other Neuroprotective Roles

However, L-serine also supports:

• Phospholipid and myelin synthesis, crucial for neuron structure
• One-carbon metabolism and methylation, which help maintain healthy brain chemistry
• Reducing cellular stress, oxidative damage, and excitotoxicity
• L-serine is the precursor to glycine. This matters because glycine is also an NMDA co-agonist (alongside D-serine). In some brain regions glycine—not D-serine—is the primary co-agonist.

So, the clinical effect might not be solely through NMDA receptor modulation.

 

Why Benefits Are Seen Mainly in Children Under 7

Several factors may explain the age effect:

1.     Brain Plasticity – Younger brains are more adaptable, so interventions may show stronger effects.

2.     Conversion to D-serine – L-serine is converted to D-serine by serine racemase, and this may be less efficient in older children.

3.     Ceiling Effects – In older children with long-standing autism and ID, neural circuits may have already stabilized in ways that make observable behavioral improvements harder.

It is unclear whether older children truly cannot benefit, or if the benefit is harder to measure with standard adaptive behavior scales.

 

Could D-Serine Directly Help Older Children?

A hypothesis is that older children might need higher levels of D-serine than their bodies can produce from L-serine. In theory:

• Direct D-serine supplementation might overcome this bottleneck.
• Safety is the main concern, as excessive D-serine can stress kidneys or neurotransmitter systems.

No large trials have tested this yet in older children with autism.

About the Researcher

Dr Yoo-Sook Joung led the AST-001 trials. She is a psychiatrist with an interest in autism interventions and has explored approaches like animal-assisted therapy. While not a basic science researcher, her clinical insights have helped design practical trials in children with severe autism and ID.

Takeaways

• High-dose L-serine shows promising results in children under 7 with severe autism and ID. The low dose was not effective.
• Benefits may involve NMDA receptor modulation, but other neuroprotective effects are likely relevant.
• Older children may require alternative approaches (e.g., D-serine), but evidence is lacking.
• Safety and careful dosing are essential; trials so far show good tolerability, with diarrhea being the most common side effect.

 

Here is the associated research leading up the recent trial

Population Pharmacokinetic Model of AST-001, L-Isomer of Serine, Combining Endogenous Production and Exogenous Administration in Healthy Subjects

AST-001 is an L-isomer of serine that has protective effects on neurological disorders. This study aimed to establish a population pharmacokinetic (PK) model of AST-001 in healthy Korean to further propose a fixed-dose regimen in pediatrics. The model was constructed using 648 plasma concentrations from 24 healthy subjects, including baseline endogenous levels during 24 h and concentrations after a single dose of 10, 20, and 30 g of AST-001. For the simulation, an empirical allometric power model was applied to the apparent clearance and volume of distribution with body weight. The PK characteristics of AST-001 after oral administration were well described by a two-compartment model with zero-order absorption and linear elimination. The endogenous production of AST-001 was well explained by continuous zero-order production at a rate of 0.287 g/h. The simulation results suggested that 2 g, 4 g, 7 g, 10 g, and 14 g twice-daily regimens for the respective groups of 10–14 kg, 15–24 kg, 25–37 kg, 38–51 kg, 52–60 kg were adequate to achieve sufficient exposure to AST-001. The current population PK model well described both observed endogenous production and exogenous administration of AST-001 in healthy subjects. Using the allometric scaling approach, we suggested an optimal fixed-dose regimen with five weight ranges in pediatrics for the upcoming phase 2 trial.

  

Population pharmacokinetic and pharmacodynamic model guided weight-tiered dose of AST-001 in pediatric patients with autism spectrum disorder

AST-001, a novel syrup formulation of L-serine, was developed for the treatment of autism spectrum disorders (ASD) in pediatric patients. This study aimed to establish a pharmacokinetic (PK)-pharmacodynamic (PD) model to elucidate the effect of AST-001 on adaptive behavior in children with ASD. Due to the absence of PK samples in pediatric patients, a previously published population PK model was used to link the PD model by applying an allometric scale to body weight. The time courses of Korean-Vineland Adaptive Behavior Scale-II Adaptive Behavior Composite (K-VABS-II-ABC) scores were best described by an effect compartment model with linear drug effects (Deff, 0.0022 L/μg) and linear progression, where an equilibration half-life to the effect compartment was approximately 15 weeks. Our findings indicated a positive correlation between the baseline K-VABS-II-ABC score (E0, 48.51) and the rate of natural progression (Kprog, 0.015 day⁻¹), suggesting enhanced natural behavioral improvements in patients with better baseline adaptive behavior. Moreover, age was identified as a significant covariate for E0 and was incorporated into the model using a power function. Based on our model, the recommended dosing regimens for phase III trials are 2, 4, 6, 10, and 14 g, administered twice daily for weight ranges of 10–13, 14–20, 21–34, 35–49, and >50 kg, respectively. These doses are expected to significantly improve ASD symptoms. This study not only proposes an optimized dosing strategy for AST-001 but also provides valuable insights into the PK-PD relationship in pediatric ASD treatment.

 

AST‐001 versus placebo for social communication in children with autism spectrum disorder: A randomized clinical trial

Aim

This study examined the efficacy of AST‐001 for the core symptoms of autism spectrum disorder (ASD) in children.

Methods

This phase 2 clinical trial consisted of a 12‐week placebo‐controlled main study, a 12‐week extension, and a 12‐week follow‐up in children aged 2 to 11 years with ASD. The participants were randomized in a 1:1:1 ratio to a high‐dose, low‐dose, or placebo‐to‐high‐dose control group during the main study. The placebo‐to‐high‐dose control group received placebo during the main study and high‐dose AST‐001 during the extension. The a priori primary outcome was the mean change in the Adaptive Behavior Composite (ABC) score of the Korean Vineland Adaptive Behavior Scales II (K‐VABS‐II) from baseline to week 12.

Results

Among 151 enrolled participants, 144 completed the main study, 140 completed the extension, and 135 completed the follow‐up. The mean K‐VABS‐II ABC score at the 12th week compared with baseline was significantly increased in the high‐dose group (P = 0.042) compared with the placebo‐to‐high‐dose control group. The mean CGI‐S scores were significantly decreased at the 12th week in the high‐dose (P = 0.046) and low‐dose (P = 0.017) groups compared with the placebo‐to‐high‐dose control group. During the extension, the K‐VABS‐II ABC and CGI‐S scores of the placebo‐to‐high‐dose control group changed rapidly after administration of high‐dose AST‐001 and caught up with those of the high‐dose group at the 24th week. AST‐001 was well tolerated with no safety concern. The most common adverse drug reaction was diarrhea.

Conclusions

Our results provide preliminary evidence for the efficacy of AST‐001 for the core symptoms of ASD.

 

The what, when and where of treating autism

The human brain is a work in progress up until your mid 20s.

It is near adult-sized at the age of 5, but many key developmental processes remain.

As brain development goes through it various steps, it requires certain genes to be activated to produce specific proteins. This is why in some single gene autisms babies are born appearing entirely typical, because at that point they are typical. Shortly thereafter when the gene cannot produce enough of its protein (haploinsufficiency) things start developing off-track. The human body is highly adaptable and the brain keeps on changing, but now on a different track.

Many dysfunctions in autism are localized to just one part of the brain and indeed you can have the opposite dysfunction in different parts of the brain at the same time. Some dysfunctions can be just transitory, or indeed just extreme in one particular developmental window.     

When it comes to NMDA activity we know that very often in autism and schizophrenia it is disturbed. But, it can be too much or too little (hyper/hypo) and very likely this changes over time and varies in different parts of the brain.

Viewed in this broader context, it is not odd to see an intervention that is most effective up to the age of seven.

  

Conclusion

If you know a child with severe autism and intellectual disability, who is under 7 years old, maybe suggest to the parents to investigate following our proactive reader Hoang and make a trial of the OTC supplement L-Serine. You can buy it inexpensively on-line, just search “L serine bulk powder.” In the US 1kg costs about $50. Just follow the dosage in the trials.

L-serine is very safe.

Using D-serine is more problematic. In clinical studies for schizophrenia and cognitive disorders, doses ranged from 30 mg/kg/day to 120 mg/kg/day in divided doses. D-serine is mostly safe at moderate doses, but very high doses carry risks of kidney stress and excitotoxicity.

Modest amounts of L-serine can be found in eggs, chicken, milk etc. The body then converts this to D-serine using an enzyme called serine racemase and vitamin B6. Once these are used up, no more D-serine can be produced “naturally.” This is why schizophrenia researchers use D-serine itself. D-serine is also sold as a bulk OTC supplement.

If the child was actually an undiagnosed Memantine-responder, you would expect to see the following if they took high dose L-serine:

·        ↑ irritability

·        ↑ sensory overload

·        ↑ hyperactivity

·        ↑ emotional volatility

·        ↑ stereotypy

·        ↑ anxiety

Because a memantine responder is a child whose biology is defined by NMDA receptor overactivity, where excessive glutamate signalling drives irritability, sensory overload, anxiety, and cognitive stress and memantine works precisely because it reduces this hyper-NMDA state.

L-serine does the opposite, it increases D-serine and so enhances NMDA activity and so in an L-serine responder it improves:

·        learning and cognitive processing

·        social attention and engagement

·        adaptive behaviour

·        overall developmental trajectory

 

In this group, the core bottleneck is not excessive glutamatergic activity but insufficient NMDA co-agonism, especially in early development when social circuits and sensory-integration networks are still forming.

 

What does “insufficient NMDA co-agonism” mean?

NMDA receptors do not work like simple on/off switches.

They need two keys to open:

·        Glutamate – the main excitatory neurotransmitter

·        A co-agonist – either D-serine or glycine

If glutamate is present but the co-agonist is missing or too low, the NMDA receptor cannot fully activate, even though the neuron is trying to fire normally.

This situation is called NMDA hypofunction caused by insufficient co-agonism

In plain terms, the glutamate system is not actually weak. The receptor is not working properly because the “second key” is missing.

 

Neural circuits needed for learning, plasticity, and social behaviour do not work properly, because the key is missing. Go find it!

   

Why does this matter in autism with ID?

Several studies (postmortem, CSF, MR spectroscopy) show that in many children with severe autism + language delay + ID, D-serine levels are reduced in key brain areas (prefrontal cortex, temporal cortex, hippocampus).

Possible reasons:

·        Low activity of serine racemase (the enzyme converting L-serine → D-serine)

·        Higher breakdown of D-serine by DAO (D-amino acid oxidase)

·        Developmentally immature astrocytes (which supply D-serine early in life)

·        Genetic factors affecting NMDA co-agonist pathways

When D-serine is low, NMDA receptors cannot activate normally even if glutamate levels are normal or high.

 

The result:

Cognitive delay, poor adaptive behaviour, weak learning reinforcement, sensory disturbances, and poor social reciprocity.

How does L-serine help?

·        L-serine is the precursor to D-serine.

 

By giving large doses of L-serine

·        The brain produces more D-serine

 

D-serine binds the NMDA co-agonist site

·        NMDA receptors can finally reach normal activation

·        Neural circuits can strengthen and rewire more effectively

·        Behaviour improves, especially in younger children where plasticity is high

 

This is why L-serine produces the opposite clinical effect of memantine:

 

• Memantine helps when NMDA activity is too high

• L-serine helps when NMDA activity is too low because of a missing co-agonist

“The treatment didn’t fail, the trial did.” Here we go again with Memantine

They are all pizza, but each slice (sub-group) is very different. Only one has pepperoni 😋 and that’s the one memantine helps!

Older regular readers will be familiar with the large, well-funded trial that started a decade ago of Memantine/Nameda for autism. It had been a widely used off-label therapy for autism in the US. The regulator asked the producer to fund a large clinical trial.

The trial failed because Memantine was shown to be no more beneficial than the placebo. That trial had 400 participants and really should have been able to identify any large sub-groups that did respond. But, it did not.

I saw that a new Memantine trial has just been published, and guess what, they found that Memantine was beneficial and they used a special kind of MRI to try and identify that sub-population.

What struck me was the type of autism population they used to make the trial. You can look at IQ, comorbid diagnosis and even what other drugs the trial subjects are already on.

Here are some stand outs:

·        The average IQ in the placebo group was 110. That is top 25% by IQ.

·        Three quarters have ADHD and multiple anxiety disorders.

·        More than half have major depression.

·        In the treatment group 20% have psychosis.

That has very little in common with my son’s presentation of “autism.”

I think it would have been better to summarise this as a trial of Aspies with ADHD, who are also likely to be very unhappy.  That is a very valid treatment group, but the word autism does not fully capture it. These could be described as “lost souls.” 

The 2025 JAMA trial shows that memantine is not a failure for autism — it’s effective for a biologically distinct, high-glutamate subgroup, often corresponding to high-IQ, emotionally dysregulated “Lost Souls” autism.

My blog was always targeted at severe types of autism, but many of the messages I receive are from people with normal to high IQ, fully verbal but deeply troubled.

You can slice the autism pizza however you like. Here are some possible slices:

·        Classic autism / Profound autism / Kanner’s autism

·        Lost souls

·        Quirky autism, Aspies, some super-brainy, but most not

·        Sub-threshold autism, self-diagnosed autism, attention-seeking diagnosis

Back to the recent trial

·        All Memantine responders had high glutamate levels, and 80% of participants with  high glutamate levels were memantine responders. 

·        Notably, the abnormally high levels of glutamate were not universal but were limited to 54.0% (n = 20 of 37) of participants with ASD 

So applying some common sense:

·        Half of “Lost souls” autism (high IQ) are likely to have high glutamate when measured in the special MRI

·        80% with this biomarker are likely to find their social impairments are reduced by Memantine. 

Hurrah !!!!  

(particularly if that Lost souls definition applies to you, or your child)

The researchers themselves summarise the results as

“In this study, treatment with memantine was superior to placebo in improving social behaviors. Youths who received memantine had 4.8 times the odds (95% CI, 1.1-21.2) of responding to treatment compared with placebo. The NNT statistic was robust (NNT = 3), indicating that 1 in 3 memantine-treated youths with ASD would respond to treatment.” 

I think the autism sample in the trial is a little odd, so I will stick to my interpretation.

If about 40% of this lost souls autism really do respond to well to 20mg a day of Memantine, why did this not show up in the 2018 trial that costs many millions of dollars? That trial excluded kids with IQ ≤70 and also required at least moderate severity of social impairment.

The IQ cut off at 70 would exclude all profound autism and about 70% of Classic/Kanner’s autism.  There really should have been plenty of responders in the 2018 trial. Was the dose too low? Somebody may have wasted $20 million in trial costs.

Memantine to Treat Social Impairment in Youths With Autism Spectrum Disorder

A Randomized Clinical Trial

Several glutamate-modulating agents, including lamotrigine, amantadine, and N-acetylcysteine, have been evaluated as potential treatments for the core symptoms of ASD, demonstrating only modest efficacy.^26-30 In contrast, preliminary data from retrospective and prospective uncontrolled trials of memantine hydrochloride, with its unique mechanism of action as a moderate-affinity noncompetitive NMDA receptor antagonist, have been promising, reporting an acceptable safety and tolerability profile and substantial improvements in SCI and RRBs in youths and adults with ASD.^31-33 However, the only controlled trial of memantine in children to date, while showing improvements in ASD behaviors, failed to demonstrate superiority over an equally robust placebo response; this is likely due to the low dosing and inclusion of participants with intellectual disability, which did not adequately assess memantine’s efficacy in individuals with ASD without intellectual disability.³⁴ Addressing these limitations, preliminary findings from an uncontrolled trial of memantine at dosages of up to 20 mg/d in adults with ASD without intellectual disability demonstrated substantial improvements in social behaviors.³⁵

Spectroscopic glutamate levels in the pgACC were significantly elevated by a large magnitude in youths with ASD compared with healthy control participants, replicating previous findings by Joshi et al²⁵ of glutamate dysregulation in individuals with ASD. Notably, the abnormally high levels of glutamate were not universal but were limited to 54.0% (n = 20 of 37) of participants with ASD, with the remainder of participants without any glutamate abnormality.

Treatment response differed based on pgACC glutamate levels in participants with ASD. A significantly greater response rate to memantine compared with placebo was observed in the high-glutamate subsample, whereas no such difference was observed in the medium-glutamate subsample. All memantine responders had high glutamate levels, and the majority of participants with ASD with high glutamate levels were memantine responders (8 of 10 [80.0%]).

 

Why is Memantine used for Alzheimer’s?

In a healthy brain, glutamate is the main excitatory neurotransmitter and is crucial for learning and memory and normal neuronal activation.

In Alzheimer’s, damaged and dying neurons start to leak glutamate into the extracellular space.
This causes chronic, low-level overactivation of NMDA receptors.

That chronic stimulation:

• lets in too much calcium (Ca²⁺)
• triggers oxidative stress and mitochondrial dysfunction
• leads to progressive neuron death — a process called excitotoxicity

Memantine’s mechanism is a partial NMDA block without shutting down normal signaling.

• It binds inside the NMDA receptor channel, only when it’s open (i.e., during overactivation).
• It blocks excessive Ca²⁺ entry, protecting neurons from excitotoxic damage.
• Because its binding is voltage-dependent and rapidly reversible, it does not block normal glutamate transmission needed for learning and memory.

Unlike strong NMDA blockers (e.g. ketamine), memantine is neuroprotective without being sedating or hallucinogenic.

Why are Alzheimer’s drugs so ineffective to treat Alzheimer’s

People start treatment for Alzheimer’s disease 25 years too late.

The biological processes start two decades before the severe symptoms appear and at that point the damage is already done.

For Memantine to be truly effective you would need to start it in your 50s.

Target people very likely to develop Alzheimer’s 25 years early

We already know who is very likely to develop Alzheimer’s. There are the 3% of the general population carrying a double copy of the risk gene, so APOE ε4/ε4.

Then we have everyone with Down syndrome. Amyloid plaques have developed by their 30s and Alzheimer’s is nowadays the leading cause of death in DS.

Why not give the option of preventative treatment? Since it is very cheap and safe.

Elevated extracellular glutamate in autism?

Many studies suggest that extracellular glutamate levels are elevated in autism, at least in certain brain regions and in subgroups of individuals.

Magnetic Resonance Spectroscopy (MRS) — often called MR Spectroscopy — is a non-invasive imaging technique that is closely related to MRI, but instead of showing brain structure, it measures brain chemistry.

MRS studies show elevated glutamate or glutamine + glutamate levels have been found in regions such as:

• Anterior cingulate cortex (ACC / pgACC)
• Basal ganglia
• Hippocampus
• Thalamus

In today’s study the biomarker for responders was elevated glutamate in the Anterior cingulate cortex (ACC / pgACC)

Excluding people from autism trials based on IQ

It is increasingly common to exclude children from autism trials based on low IQ.

The right approach is probably to have separate trials for those with IQ less than 80. Then you can adapt the assessment process to suit people with limited communication and cognitive skills.

As more and more people want to get diagnosed with level 1 autism, the relevance of those at level 3 tends to get minimized. This does upset many parents.

I actually believe that raising IQ should be an endpoint in some autism trials. It is meaningful and measurable. I was very surprised to discover that you can raise IQ. Raising IQ by just 10 points is a big deal in classic/profound autism, it then makes all that expensive 1:1 therapy in childhood much more effective.

Conclusion

Memantine is a really cheap Alzheimer’s drug that can help some autistic people with high glutamate in their brain. It has been shown to be very safe.

There are much more expensive drugs like Riluzole.

Riluzole acts on multiple glutamate-related pathways:

·        ↓ Glutamate release from presynaptic neurons

·        ↑ Glutamate reuptake by astrocytes (enhancing EAAT2 transporter activity)

·        Blocks voltage-dependent sodium channels, reducing neuronal overactivation

·        Modulates NMDA and AMPA receptors, dampening excessive excitatory signaling

This results in lower extracellular glutamate, less calcium influx, and protection from oxidative stress and excitotoxicity — a key mechanism in ALS, Alzheimer’s, and possibly some autism.

I did try both Memantine and Riluzole a long time ago. Riluzole caused lethargy, which makes sense.

In Memantine clinical trials, social responsiveness is the measure the researchers always like to focus on.

Some people think the big expensive 2015-2018 trial had the wrong primary endpoint — Memantine might have modestly improve cognitive flexibility or irritability, but those were secondary measures.

The cost of the special MRI scan to diagnose high glutamate in the PgCC (Pre-genual Anterior Cingulate Cortex) is $1000 to $2000. This is way too expensive to be used at a wide scale as a biomarker.

20mg a day of Memantine costs $20 a month in the US and even less everywhere else.

Peter’s approach would be: “don’t overthink it, just try it!”

Do not expect it to be a silver bullet. If it helps, add it to your polytherapy. If there is no positive response, stop the therapy. A good investment of $20, either way.