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.