Looking at the Yale perspective on identifying therapies for the downstream effects of a spectrum of autism genes

  

 

Our reader Aleksandra recently sent me a complicated paper by Meilin Fernandez Garcia and colleagues from Yale. The corresponding senior author is Professor Kristen Brennand, whose laboratory is internationally recognised for using human induced pluripotent stem cells (hiPSCs), CRISPR gene editing and neuronal models to study neuropsychiatric disorders.

At least one autism parent is doing something similar to help their child. Instead of engineering mutations into generic cell lines to identify therapies that might benefit many patients, the same approach is being applied to neurons derived from a single individual with autism, allowing researchers to search for drugs that reverse that person's unique cellular abnormalities. You do need money to this.

 

Transcriptomicand phenotypic convergence of neurodevelopmental disorder risk genes in vitroand in vivo

Diverse risk genes have been identified for neurodevelopmental disorders (NDDs), but how these genes converge on similar biological pathways in neurons, and thus give rise to similar phenotypes, is unclear. Here we apply a pooled CRISPR approach to successfully target 23 NDD loss-of-function genes with roles in chromatin biology and examine convergent effects on gene expression across human induced pluripotent stem cell-derived neural progenitor cells, glutamatergic neurons and GABAergic neurons. Points of convergence vary between these cell types, with the greatest number of convergent genes and strongest convergent networks in mature glutamatergic neurons, where they broadly represent synaptic, epigenetic and, unexpectedly, mitochondrial pathways. The most convergent networks were observed between NDD genes with shared biological annotations, clinical associations and co-expression patterns in human post-mortem brain. Drugs that were predicted to reverse convergent transcriptomic signatures and/or arousal and sensory processing behaviors ameliorated behavioral phenotypes in zebrafish NDD gene mutants. These results suggest that convergent effects of NDD risk genes could provide clinically useful insights.

 

As I looked through this paper, I experienced an unusual feeling. The methodology could hardly be more different from the approach I have taken over the past decade, yet many of the conclusions felt surprisingly familiar.

This blog has never had research grants, laboratories, graduate students or sophisticated genomic platforms. It has simply been the product of many years of reading the scientific literature, following developments in genetics, pharmacology and physiology, and trying to understand how apparently unrelated findings might fit together.

The Yale team approached the problem from the opposite direction. Using CRISPR gene editing, single-cell transcriptomics, machine learning and computational drug discovery, they asked a simple question:

Do hundreds of different neurodevelopmental disorder genes converge on a smaller number of common downstream biological pathways that might be therapeutically targetable?

Their answer was yes.

For long-time readers of EpiphanyASD, that conclusion will sound very familiar.

From hundreds of autism genes to common biology

One of the recurring themes of this blog has been that autism is unlikely to require hundreds of completely different treatments simply because there are hundreds of different autism genes.

Many genetic mutations disrupt the same biological systems.

If those downstream pathways can be identified and safely modulated, treatment may become considerably simpler than developing a separate therapy for every mutation.

The Yale investigators reached this conclusion using an extraordinarily sophisticated experimental pipeline.

They used CRISPR to knock out 23 neurodevelopmental disorder genes in human stem-cell-derived neurons and performed single-cell RNA sequencing on more than 118,000 cells. Machine-learning models were then used to extend these findings to more than one hundred neurodevelopmental disorder genes.

Instead of focusing on what made each mutation different, they searched for what they had in common.

The strongest convergence occurred in mature glutamatergic neurons, where many mutations disrupted shared transcriptional programmes involving synaptic function, chromatin regulation and, interestingly, mitochondrial biology.

 

Using AI to identify potential therapies

The next stage of the study was particularly elegant.

Rather than beginning with drugs already proposed for autism, the researchers searched the Connectivity Map database for medicines that produced gene-expression changes opposite to those caused by the convergent neurodevelopmental signatures.

In effect they asked:

Which existing drugs make diseased neurons look genetically more like healthy neurons?

After computational ranking, the most promising candidates were tested in zebrafish models.

Ten of the eleven drug–gene combinations rescued at least one behavioural phenotype.

That does not demonstrate efficacy in people with autism, but it does provide strong proof-of-principle for this approach.

 

Looking beyond the individual drugs

Being a practical sort of person, my first instinct was to examine the supplementary data rather than simply read the discussion section.

With the help of AI, this took only a few minutes.

The Yale analysis compared approximately 520 drug signatures from the Connectivity Map against the convergent transcriptomic changes. Although this represents only a fraction of all approved medicines, it is sufficient to identify the biological pathways most strongly associated with transcriptomic reversal.

I was curious to see how these pathways compared with those that have gradually emerged on EpiphanyASD over more than ten years.

The comparison encouraged me.

Biological Pathway	Yale Computational Hits	Epiphany ASD Approach	Interpretation
Mevalonate / Cholesterol	Pravastatin, Rosuvastatin	Atorvastatin	Independent identification of statin therapy, although different statins were selected.
Renin–Angiotensin System	Valsartan	Telmisartan	Different ARBs targeting the same signalling pathway.
Calcium Signalling	Diltiazem	Verapamil, Amlodipine	Different L-type calcium-channel blockers acting on related pathways.
Histamine / Remyelination	Clemastine	Clemastine	Complete agreement.
mTOR Signalling	Sirolimus	—	Pathway highlighted by Yale and EpiphanyASD. No safe drug.
Immune / Inflammatory Modulation	Corticosteroids	Pioglitazone, NAC	Different therapeutic strategies aimed at immune regulation.
Chloride Homeostasis	—	Bumetanide	Proposed by Ben-Ari and adopted by EpiphanyASD but not identified by the transcriptomic analysis.
Oxidative Stress / Mitochondrial Support	—	NAC, Alpha-lipoic acid, Taurine	Important focus of this blog but not highlighted by the Yale analysis.
Polyamine / NMDA Modulation	—	Agmatine	Outside the principal pathways identified by Yale.

The important point is not that the same drugs were identified.

They were not.

Nor does this study validate any treatment combination discussed on this blog.

The interesting observation is that two completely independent approaches have converged on many of the same biological pathways.

One approach relied on years of reviewing genetics, pharmacology, physiology and clinical reports.

The other relied on CRISPR gene editing, human neuronal models, single-cell transcriptomics, machine learning and computational drug discovery.

When such different methodologies begin pointing towards the same pathways, it increases confidence that those pathways deserve serious attention.

 

Why this matters

One feature of the Yale study particularly resonated with me.

The computational analysis did not identify a single "autism drug."

Instead, it identified multiple interacting biological systems. This mirrors another long-standing theme of EpiphanyASD.

If autism involves disturbances in several interconnected biological systems, then it seems unlikely that one drug alone will restore normal function in many patients.

Instead, carefully chosen combinations of therapies, each targeting a different aspect of the biology, may ultimately prove more effective than monotherapy.

That remains a hypothesis requiring proper clinical evaluation, but the Yale analysis certainly does not argue against such an approach.

 

Where the approaches differ

The Yale study identified sirolimus as an interesting candidate, whereas this has not previously featured prominently on this blog. Some readers have trialed it, but there are side effects, so it is not high on my list of practical therapies.

Conversely, therapies frequently discussed here—including bumetanide, NAC, alpha-lipoic acid, taurine and agmatine—did not emerge from the transcriptomic analysis.

That should not be overinterpreted.

The Connectivity Map used in this study analysed approximately 520 compounds, representing only a subset of approved medicines. Closely related drugs such as atorvastatin, telmisartan and verapamil were either absent from the analysed library or did not emerge as leading transcriptomic reversers.

Transcriptomic reversal is only one method for identifying therapeutic candidates. In the case of a very complex conditional like autism it can likely only solve part of the problem.

Transcriptomic mapping captures gene expression reversal, whereas therapies like Bumetanide (NKCC1 chloride transporter) or NAC (direct antioxidant/glutamate buffer) act primarily through direct protein/ion channel activity, which might not generate a strong early transcriptomic signature in cell culture models.

 

What proportion of effective drugs would be missed by the Yale approach?

 

The 3-Bucket Pharmacological Model

When evaluating how small-molecule drugs work across all FDA-approved therapeutics—and specifically within central nervous system (CNS) and neurodevelopmental disorders—drugs fall into three distinct functional buckets based on their primary Mechanism of Action (MoA):

 

 

A transcriptomic-only screening platform (scRNA-seq + Connectivity Map) is structurally blind to Buckets 2 and 3, meaning it may miss approximately 50% to 60% of effective CNS and neurodevelopmental therapeutics.

 

Breakdown by Bucket

Bucket 1: Primary Transcriptomic & Genomic Modulators (about 40% of drugs)

• Mechanisms: Nuclear receptor agonists/antagonists, chromatin remodelers, HDAC inhibitors, and upstream growth factor/kinase cascades (e.g., mTOR signaling).
• Representative Therapies: Corticosteroids, Pioglitazone (PPAR-gamma), Sirolimus (mTOR), Statins, and ARBs.
• Screening Sensitivity: HIGH. Because these compounds work by intentionally altering mRNA transcription and gene expression programs, scRNA-seq and CMap pick them up effectively. This is where the Yale pipeline excels.
•  

Bucket 2: Electrophysiological & Membrane Modulators (about 40% of drugs)

• Mechanisms: Direct blockade or opening of ion channels, cell-membrane transporters, and fast ligand-gated ion channels. These alter intracellular ion concentrations and membrane voltage within milliseconds to minutes.
• Representative Therapies: Bumetanide (NKCC1 chloride-transporter blocker), Calcium channel blockers (Diltiazem, Verapamil), GABA-A allosteric modulators, and Sodium channel blockers.
• Screening Sensitivity: LOW (High Risk of Being Missed). The primary therapeutic event is electrophysiological (shifting excitation/inhibition balance or hyperpolarizing neurons). While long-term cellular adaptation can produce secondary transcriptional noise, the primary mechanism produces little to no acute transcriptomic signature in cell culture.
•  

Bucket 3: Direct Metabolic & Biochemical Buffers (about 20% of drugs)

• Mechanisms: Direct stoichiometric chemical reactions, free-radical scavenging, direct precursor supply for endogenous enzymatic cycles, or polyamine/receptor modulation that occurs without altering target gene expression.
• Representative Therapies: N-Acetylcysteine (NAC) (direct ROS scavenger and glutathione precursor), Alpha-Lipoic Acid, Taurine, and Agmatine (polyamine/NMDA pathway modulator).
• Screening Sensitivity: VERY LOW (Almost Entirely Missed). These molecules correct cellular redox potential or metabolic fluxes directly. Because they act via direct biochemistry rather than transcriptional reprogramming, a gene expression screen will almost always drop them.

 

Why the "miss rate" is exceptionally high in Autism / NDDs

In fields like oncology, where the goal is altering cell cycle or cell death pathways, Bucket 1 dominates (making transcriptomic matching highly effective).

However, in Neurodevelopmental Disorders (NDDs), core neurobiology relies heavily on:

1.     Intracellular ionic gradients & E/I balance (Bucket 2)

2.     Mitochondrial health, redox balance & oxidative stress (Bucket 3)

Because roughly 50-60% of potential NDD interventions live in Buckets 2 and 3, relying solely on transcriptomic reversal naturally drops more than half of the potential drugs.

To capture those therapies, systems biology must complement transcriptomic screens with Microelectrode Arrays (MEAs) for Bucket 2 and Functional Metabolomics/High-Content Imaging for Bucket 3.

  

Final thoughts

For me, this paper is encouraging for two reasons.

First, it provides compelling evidence that many genetically distinct neurodevelopmental disorders converge on a relatively small number of shared downstream biological pathways.

Second, after more than a decade of independent, unfunded research on EpiphanyASD, it is gratifying to see that a large, well-funded research programme at one of the world's leading universities has independently highlighted many of those same pathways and drugs.

That does not prove that any particular treatment discussed on this blog is correct.

It does suggest that the central idea—that autism should often be approached as a disorder of convergent downstream biology rather than hundreds of isolated gene defects—is increasingly supported by modern systems biology.

If future studies continue to point in the same direction, the prospect of rational, pathway-based precision medicine for autism becomes much more realistic.

 

One final thought. Looking back at my original TRH hypothesis from more than a decade ago, today's technology finally offers a practical way to test it. Rather than immediately embarking on a clinical trial, a logical first step would be to investigate both the direct TRH super-agonist taltirelin (Ceredist) and the indirect TRH-enhancing approach using rifaximin in patient-derived induced pluripotent stem cells (hiPSCs), differentiated into neurons.

The rifaximin story is particularly intriguing because it appears to stimulate endogenous TRH production through the gut–brain axis, rather than replacing TRH pharmacologically. This is exactly the type of experimental platform now being used by groups such as Professor Kristen Brennand's laboratory at Yale to identify drugs that reverse disease-related transcriptomic abnormalities. If either taltirelin or rifaximin were shown to normalize gene-expression patterns or neuronal function in patient-derived neurons, it would provide a much stronger scientific rationale for moving on to carefully designed clinical studies. It would also be fascinating to compare the effects of the two approaches directly, asking whether a TRH super-agonist and a gut-mediated increase in endogenous TRH ultimately converge on the same downstream molecular pathways. If they did, it would represent another example of two completely different therapeutic strategies converging on the same underlying biology—a recurring theme throughout this article

 

Epiphany: TRH and Rifaximin – an alternative to intranasal TRH or oral Taltirelin/Ceredist?

Epiphany: The Peter Hypothesis of TRH-induced Behavioural Homeostasis in Autism

 

Taltirelin sits directly in Bucket 1 as a G Protein-Coupled Receptor agonist that triggers immediate nuclear gene transcription (upregulating BDNF etc), whereas Rifaximin overlaps Buckets 1 and 3 by using gut-microbiome metabolic modulation to indirectly drive central TRH expression via the gut–brain axis.

 

 

Beyond Profound Autism - Towards More Useful Ways to Classify Autism

 

A new paper continues to refine the concept of Profound Autism to define how should we describe autistic people who require lifelong, round-the-clock care. 

Developing a consensus research definition for profound autism using a modified Delphi method

 

Using a modified Delphi process involving researchers, clinicians, caregivers and autistic individuals, the authors developed a consensus research definition of profound autism. Their goal is admirable. By encouraging researchers to use a common definition, studies become easier to compare and the subgroup with the greatest support needs is less likely to be overlooked.

I welcome that goal, but …

 

Research definition or clinical reality?

The authors are careful to emphasise that "profound autism" is intended as a research definition rather than a new clinical diagnosis.

That distinction is important.

Researchers need consistent definitions so that studies include comparable participants. A research definition helps ensure that one study of "profound autism" is investigating a similar population to another.

Parents, however, have often been asking for something completely different.

Many families caring for autistic people who require lifelong support argue that the autism spectrum has become so broad that it no longer communicates the reality of their child's disability. They are looking for terminology that helps explain prognosis, support needs and, in some countries, eligibility for services.

A research-only definition does not solve those problems.

Indeed, by combining autism severity, intellectual disability and adaptive functioning into a single category, it may create new sources of confusion outside the research setting.

 

This is in marked contrast to Asperger's syndrome. In 1994, Asperger's syndrome was introduced into the DSM-IV as a recognised clinical diagnosis, not merely as a research category. It was used by clinicians, educators and families to describe a distinctive presentation of autism. When it was removed in 2013 and absorbed into the broader diagnosis of Autism Spectrum Disorder, the spectrum expanded to encompass individuals ranging from those living entirely independently to those requiring lifelong, round-the-clock care.

It is therefore unsurprising that researchers are once again looking for ways to subdivide this very broad spectrum.

However, unlike Asperger's syndrome, profound autism is not intended to become a clinical diagnosis. It is a tool for research, not for clinical practice.

Perhaps what is really needed are two complementary frameworks:

·        A research framework that allows scientists to compare studies using consistent and well-defined populations.

·        A clinical framework that describes an individual's autism severity, intellectual ability, adaptive functioning, language, medical complications and support needs separately.

The second approach may be far more useful for families because it paints a richer and more accurate picture of the individual than any single label ever could.

The authors deserve credit for recognising that the current autism spectrum has become too broad for many research questions and their perseverance. Whether "profound autism" is the best way to address that problem remains open to debate, but they have undoubtedly started an important conversation.

How Has the Diagnosis Changed?

It is easy to forget how much the diagnosis of autism has changed over the past quarter of a century.

In 2000, clinicians were still using the DSM-IV classification, which recognised several distinct pervasive developmental disorders: Autistic Disorder, Asperger's syndrome and PDD-NOS (Pervasive Developmental Disorder – Not Otherwise Specified).

To receive a diagnosis of Autistic Disorder, a child had to meet at least six diagnostic criteria spanning social interaction, communication and restricted or repetitive behaviours, with symptoms present before three years of age. Most children diagnosed with Autistic Disorder had significant language delay, many had intellectual disability, and a large proportion required substantial lifelong support.

By contrast, individuals with average or above-average intelligence and no clinically significant language delay were generally diagnosed with Asperger's syndrome, while those who showed autistic features but did not fully satisfy the criteria for either diagnosis often received a diagnosis of PDD-NOS.

This changed dramatically in 2013 when DSM-5 replaced these separate diagnoses with a single diagnosis of Autism Spectrum Disorder (ASD). The intention was to improve diagnostic consistency, recognising that there were no clear biological boundaries separating the previous categories. However, the consequence was the creation of an exceptionally broad diagnostic spectrum.

Today, the same diagnosis encompasses people living entirely independently—including university professors, engineers, scientists and other professionals—as well as individuals who cannot communicate verbally, require assistance with every aspect of daily living and need lifelong, round-the-clock care.

This historical perspective helps explain why researchers are once again attempting to identify meaningful subgroups within autism. In many respects, the proposal for profound autism represents an effort to recover some of the distinctions that were lost when Asperger's syndrome and PDD-NOS were merged into a single spectrum. 

Why do we need a better classification of autism?

Autism has become an extraordinarily broad diagnosis.

At one end of the spectrum are university professors, software engineers, airline pilots and people who require little or no formal support.

At the other end are individuals who cannot communicate verbally, cannot dress or feed themselves, may have epilepsy, severe gastrointestinal disease, self-injurious behaviour and require lifelong supervision.

Although these individuals all satisfy the behavioural criteria for autism spectrum disorder, they clearly do not represent the same clinical condition.

When research studies combine such diverse populations under one diagnostic label, interpreting the results becomes increasingly difficult. A treatment that benefits one subgroup may have no effect—or even be harmful—in another.

Recognising this enormous heterogeneity is one of the greatest strengths of the new paper.

 

The proposed definition of profound autism

After two rounds of expert consultation, the Delphi panel proposed that profound autism should be defined by several key characteristics.

The individual should:

• meet the diagnostic criteria for autism spectrum disorder
• be at least eight years old
• require continuous adult supervision to ensure health, safety and wellbeing
• demonstrate adaptive functioning well below age expectations

and also have either:

• severe intellectual impairment (typically reflected by an IQ below 50), or
• minimal verbal communication, consisting mainly of single words or fixed phrases used to communicate basic needs.

Importantly, this is intended as a research definition, not a new clinical diagnosis.

The intention is not to create another subtype of autism but to ensure that studies investigating those with the highest support needs are describing comparable populations.

That is an entirely reasonable objective.

 

What the Paper Gets Right

Several aspects of the paper deserve praise.

First, it openly acknowledges that autism is not one disorder.

The current autism spectrum encompasses people with enormously different abilities, medical problems, educational outcomes and support requirements.

Recognising that these individuals should not always be analysed together is long overdue.

Secondly, the authors move away from relying solely on IQ.

Instead, they give much greater emphasis to adaptive functioning—the practical skills needed for everyday life.

Being able to dress, wash, communicate needs, recognise danger, prepare food or manage daily routines often determines independence far more than an IQ score.

In many ways, adaptive functioning is a better measure of the help a person actually requires.

Finally, the authors recognise that research involving those with the greatest support needs has declined over recent decades.

That imbalance deserves to be corrected.

 

Where I think the definition falls short

Although I agree with the motivation behind the paper, I am less convinced by the solution.

My concern is that the proposed definition combines three different concepts into a single label:

• autism severity
• intellectual disability
• adaptive functioning

These are related, but they are not identical.

A person may have extremely severe autistic symptoms while having an IQ of 120.

Another may have equally severe autistic symptoms together with profound intellectual disability.

A third may have only moderate autistic features but such poor adaptive functioning that lifelong supervision is still required.

These individuals have very different developmental trajectories, medical needs and treatment priorities.

Grouping them together under one behavioural label risks obscuring those important differences.

 

James Coplan's alternative framework

More than twenty years ago, pediatric neurologist James Coplan proposed a remarkably elegant way of thinking about autism prognosis. I wrote about it in my book on page 90.

Instead of asking a single question—

"How severe is the autism?"

—he separated prognosis into independent dimensions.

His framework considered:

• severity of autistic features
• intellectual ability
• age

 

 

 

 

 

His diagrams illustrate two children who initially appear equally autistic.

Child A has profound intellectual disability.

Child B has normal intelligence.

Both improve with age and both still have autism. But child B’s intelligence allows him to break up the troubling aspects of autism into smaller more manageable pieces. Child A never manages to overcome these barriers and remains disabled.

Their long-term outcomes become completely different.

The difference is not autism alone.

It is autism combined with intellectual disability.

I have always thought this framework was one of the clearest ways of explaining why prognosis varies so enormously across the autism spectrum.

If I were updating Coplan's model today, I would start with a fourth independent dimension, adaptive functioning. Later we can add some additional dimensions to narrow down sub-groups of autism.

 

 

 

 

Autism severity is not the same as IQ

One recurring source of confusion is the assumption that severe autism automatically means low intelligence. It does not.

Many autistic people experience severe sensory difficulties, severe social disability and marked repetitive behaviours while having average, or even exceptionally high, IQ.

Conversely, many autistic people also have severe intellectual disability.

These are different dimensions.

This distinction matters because one reason the term "profound autism" has generated debate is that many autistic people with severe autistic symptoms but normal intelligence feel it reinforces the misconception that intellectual disability is simply part of autism, rather than a common but separate co-occurring condition.

The authors of the Delphi paper were clearly aware of this problem.

Interestingly, the expert panel could not reach consensus on the most appropriate IQ threshold.

An IQ below 50 received the greatest support but still failed to reach the predefined level required for formal consensus.

That illustrates just how difficult it is to reduce a complex clinical picture to a single number.

 

The taboo of low IQ

One recurring issue when discussing autism today is the reluctance to talk openly about impaired cognitive function, or low IQ.

This is not an entirely new phenomenon. In 1973 the IQ threshold traditionally associated with what was then called mental retardation (now intellectual disability) became more stringent. Whereas an IQ below about 80 had been used previously, the emphasis shifted to an IQ below 70, approximately two standard deviations below the population mean. Although this appears to be a relatively small numerical change, it reduced the proportion of the population meeting the criterion from roughly 16% to about 2.3% today.

Today we seem to be moving in opposite directions. The diagnostic boundaries for autism and ADHD have broadened considerably, while the definition of intellectual disability become narrower.

Unfortunately, IQ has become an uncomfortable subject. Rather than asking why some children have impaired cognitive development—and whether it might be prevented or treated—we often focus on finding more acceptable terminology, such as intellectual disability or learning difficulty. Changing the name may reduce stigma, but it does not change the underlying biology.

IQ is not determined by genetics alone. It is influenced by prenatal health, nutrition, education, environmental factors and many medical conditions. The remarkable increase in average IQ observed in countries such as South Korea over the past several decades demonstrates that cognitive performance can improve substantially across a population as living conditions, health and education improve.

South Korea provides one of the world's clearest examples of how environmental factors can influence measured cognitive ability. During the country's rapid economic development, researchers documented exceptionally large gains in IQ test performance—around 7–8 points per decade for cohorts born between 1970 and 1990. These gains are far too rapid to be explained by genetics and instead reflect the powerful effects of improved nutrition, healthcare, education and childhood development.

Average IQ in sub-Saharan Africa is reported to be around 80, but it is not a politically correct subject to discuss.

Perhaps the most encouraging development is the growing recognition that some forms of intellectual disability are treatable. This indeed ended up being one of the goals of my son’s autism therapy.

I was surprised to later discover the excellent Treatable-ID.org project, which catalogues hundreds of inherited metabolic and genetic disorders in which early diagnosis and targeted treatment can prevent or reduce intellectual disability. The database is not yet complete, but it highlights how understanding the biology of cognitive impairment can lead to meaningful interventions.

Parents often face a difficult dilemma. When applying for educational or social support, they may need to provide evidence that their child has intellectual disability. Yet, understandably, they also want to emphasise their child's strengths and abilities. That tension can make honest discussions about cognitive ability surprisingly difficult.

Ultimately, our goal should not simply be to find kinder terminology. It should be to understand why intellectual disability develops and, wherever possible, to prevent or treat it.

 

CARS, IQ and Adaptive Functioning Measure Different Things

One lesson from this discussion is that we should stop expecting a single measurement to describe everything.

Autism researchers already have well-established tools for measuring different aspects of an individual's condition.

The Childhood Autism Rating Scale (CARS) is widely used in clinical trials to quantify the severity of autistic symptoms. It allows researchers to determine whether a treatment has reduced the core behavioural features of autism.

IQ tests measure cognitive ability.

Adaptive behaviour scales, such as the Vineland Adaptive Behavior Scales, measure how well a person functions in everyday life.

Language assessments measure communication.

These are not competing measurements.

They answer different questions.

 

Question	Measurement Tool	What It Isolates
How severe are the autistic features?	CARS (Childhood Autism Rating Scale)	Core behavioral features & symptom severity
How well does the brain solve problems?	IQ Tests	Cognitive and intellectual ability
How independently can the person function?	Vineland (Adaptive Behavior Scales)	Practical, real-world everyday life skills
How well can they communicate?	Language Assessment	Structural and functional communication levels

 

No single score can replace the others.

For example, two children may both have a CARS score of 46, indicating similarly severe autistic symptoms.

One may have an IQ of 110.

The other an IQ of 35.

Their autism severity is similar.

Their prognosis is not.

Likewise, two people with identical IQs may have very different adaptive functioning. One may travel independently and prepare meals, while another may require constant supervision because of poor judgement or inability to communicate in dangerous situations.

Each measurement adds a different piece of the puzzle.

Although adaptive functioning may ultimately predict support needs better than IQ, IQ remains a concept that most people readily understand. Saying that someone has an IQ below 50 immediately conveys severe cognitive impairment. Saying that adaptive functioning is "three standard deviations below the mean" is scientifically precise but far less intuitive.

For parents, describing an eight-year-old as having adaptive skills comparable with those of an eighteen-month-old child may actually communicate the situation more clearly, even though psychologists generally avoid age-equivalent scores because of their statistical limitations.

Rather than trying to compress all of these dimensions into a single label, we should recognise that each contributes valuable information.

 

What About Self-Injury and Aggression?

One of the most surprising findings from the Delphi process is not what was included—but what was deliberately excluded.

The expert panel decided that self-injurious behaviour (SIB), aggression and other severe challenging behaviours should not form part of the definition of profound autism.

Their reasoning is understandable.

Although these behaviours are common among autistic people requiring lifelong care, they are not universal. Some profoundly disabled autistic people never become aggressive or self-injurious, while others with less severe intellectual impairment do.

From the perspective of defining a research population, that makes sense.

However, from the perspective of improving lives, I think something important risks being overlooked.

Self-injury and severe aggression are among the greatest causes of:

• family stress
• school exclusion
• psychiatric admission
• physical injury
• use of restrictive practices
• reduced quality of life

These behaviours are not simply unfortunate complications.

They are major clinical problems in their own right.

They also almost certainly represent important biological processes that deserve investigation.

Whether driven by pain, epilepsy, anxiety, executive dysfunction, sensory overload, catatonia, immune activation or other mechanisms, severe behavioural dysregulation should not disappear from autism research simply because it does not define a diagnostic subgroup.

Instead, it should become a major research priority.

 

Towards a Multidimensional Framework

Rather than inventing another umbrella label, I believe autism research would benefit from routinely describing individuals across several independent dimensions.

For example:

Dimension	Example measure
Autism severity	CARS-2
Intellectual ability	IQ
Adaptive functioning	Vineland
Language	Single words, phrases, fluent speech
Behaviour	Self-injury, aggression, catatonia
Medical complexity	Epilepsy, gastrointestinal disease, sleep disorder
Development	Age and developmental trajectory

 

This immediately creates a far richer description than any single label.

Instead of saying:

"This study included children with profound autism."

Researchers could report:

• Mean CARS score
• Mean IQ
• Mean adaptive functioning
• Language level
• Prevalence of epilepsy
• Frequency of self-injury
• Frequency of aggression
• Presence of regression
• Known genetic diagnoses

Readers would immediately understand precisely which population had been studied.

 

 

 

Then, for example, we have some Aspies with IBD (irritable bowel disease). The IBD does share a biological basis with their autism. Treat the IBD and watch their autism symptoms mellow.

The future lies in studying better subgroups

Perhaps the greatest opportunity is not to invent another autism label, but to identify clinically and biologically meaningful subgroups.

Many of these already exist.

Examples include:

• Autism with profound intellectual disability
• Autism with average or high intelligence
• Autism with severe self-injury
• Autism with severe aggression
• Autism with epilepsy
• Autism with catatonia
• Autism with developmental regression
• Autism associated with mitochondrial dysfunction
• Autism with severe gastrointestinal disease
• Autism with chronic sleep disorders
• Autism associated with immune dysregulation
• Autism following hypoxic brain injury
• Autism following congenital infection
• Autism associated with connective tissue disorders
• Autism associated with specific genetic syndromes such as SCN2A, SHANK3, CACNA1C, SCN8A, SYNGAP1, Phelan-McDermid syndrome, Rett syndrome, Fragile X syndrome, Tuberous Sclerosis Complex and many others.

 

Even these subgroups are not mutually exclusive.

A child may belong to several simultaneously.

That is perfectly acceptable.

The purpose is not to create dozens of new diagnoses.

The purpose is to identify people who are likely to share similar biology—and therefore similar treatment opportunities.

Interestingly, autism research is already moving in this direction.

Clinical trials increasingly recruit specific subgroups rather than "all autism."

Examples include:

• minimally verbal autism
• autism with epilepsy
• Fragile X syndrome
• Rett syndrome
• tuberous sclerosis
• Phelan-McDermid syndrome
• SCN2A-related autism

The field has already recognised that autism is biologically heterogeneous.

Perhaps our classification systems should catch up.

 

Precision medicine needs better descriptions

The real promise of precision medicine is not simply to identify ever smaller genetic subgroups.

It is to understand why individuals who satisfy the same behavioural diagnosis often have completely different biology.

One child may have mitochondrial dysfunction.

Another an ion channel disorder.

Another chronic neuroinflammation.

Another an underlying metabolic disease.

Another significant prenatal hypoxic brain injury.

Behaviour alone cannot distinguish these individuals.

Yet these biological differences may determine prognosis far more accurately than behavioural scores.

They may also determine which treatments are most likely to work.

The future of autism research therefore lies not in inventing broader behavioural labels but in combining behavioural assessment with genetics, neurobiology, metabolism, electrophysiology and careful clinical phenotyping.

That is the essence of precision medicine.

 

Conclusion

The new consensus definition of profound autism represents an important step forward.

It recognises that autism encompasses individuals with enormously different abilities and support needs, and that research should no longer treat them as one homogeneous population.

I fully support that objective.

However, I am not convinced that another behavioural label is the final answer.

Autism severity, intellectual ability, adaptive functioning, language, challenging behaviour and medical complexity are all independent dimensions.

They should be measured independently.

James Coplan recognised this more than twenty years ago when he separated autism severity from intellectual ability and developmental trajectory.

Today I would simply extend that framework by adding adaptive functioning, language, behaviour and medical complexity.

Most importantly, I believe the future lies beyond behavioural classification altogether.

The greatest advances in autism research are increasingly coming from the identification of biologically meaningful subgroups—whether defined by genetics, epilepsy, mitochondrial dysfunction, immune abnormalities, catatonia, regression or other underlying mechanisms.

These subgroups are far more likely to reveal targeted treatments than any single behavioural label.

The future of autism research is therefore unlikely to be shaped by a new definition such as profound autism.

It will be shaped by understanding why two people with the same autism diagnosis can have completely different biology, completely different outcomes and, ultimately, completely different treatments.

Labels describe people. Biology explains them.

 

The average IQ of South Koreans has risen dramatically over the past half century, largely through improvements in nutrition, healthcare and education. Whether the same can be said for autism researchers is perhaps more debatable. In our enthusiasm to develop new classifications, we should not overlook the valuable work carried out decades ago. James Coplan's multidimensional framework remains remarkably relevant today and arguably aligns more closely with the goals of precision medicine than many more recent attempts to subdivide the autism spectrum. Hans Asperger was 50 years ahead of his time.

 

 

Note on Hans Asperger

A 2023 investigation concluded that there was no evidence Hans Asperger knew that Am Spiegelgrund was operating a child euthanasia programme when he referred patients there in 1941. Around 800 children were ultimately killed at the institution during the Nazi period, including two of Asperger's former patients.

https://pubmed.ncbi.nlm.nih.gov/36239413/

Following the end of the Second World War, Ernst Illing, the director of Am Spiegelgrund, was arrested and tried by an Austrian court in 1946. He was convicted of murders committed at the institution, sentenced to death, and executed by hanging on 30 November 1946.

The historical record surrounding Hans Asperger continues to be debated. New evidence should always be considered carefully, but historical interpretations should be revised only when supported by robust evidence. Replacing one oversimplified narrative with another risks introducing new biases rather than improving our understanding of the past.

It is striking that some people are deeply concerned about the former diagnosis of Asperger syndrome because of ongoing historical debates about Hans Asperger and yet many give little thought to buying products from companies with well-documented links to the Nazi regime, such as Hugo Boss, whose founder was a Nazi Party member and whose company manufactured uniforms for Nazi organisations while using forced labour during the Second World War.  

In 1938, the Nazi government even awarded Henry Ford the Grand Cross of the German Eagle, the highest honour it could bestow on a foreign civilian.

Henry Ford was a pioneering industrialist, but he also promoted antisemitic views through his newspaper, the Dearborn Independent. These articles were later collected into the four-volume series The International Jew in the early 1920s. It was translated into at least 12 languages. The writings spread antisemitic conspiracy theories about Jewish people and influenced extremist movements internationally, including Nazi Germany. Adolf Hitler openly admired Ford and mentioned him favourably in Mein Kampf. Ford was the only American mentioned positively by Hitler in Mein Kampf, and Hitler later stated that he regarded Ford as an inspiration.

It does illustrate that society often applies historical scrutiny inconsistently.

I hope Simon Baron-Cohen never bought a Ford!

Is it relevant today? Not to most people.