What ASH1L models can—and cannot—tell us

One finding that changed my own thinking was that the consequences of ASH1L loss did not appear simply as a fixed abnormality established early in development. In our mouse studies, neuronal hyperactivity was not prominent at an early postnatal stage but became evident later, with increased neuronal activity observed in multiple brain regions at P30 and P60. The mice also showed increased susceptibility to induced seizures and persistent locomotor hyperactivity.

These observations made me reconsider a relatively simple model in which loss of ASH1L produces an early developmental defect that then remains essentially unchanged. I now think it may be more useful to view ASH1L deficiency as altering a developmental trajectory. Early disruption of chromatin regulation and gene expression may affect how neural cells differentiate, mature, and establish circuits. Some abnormalities may emerge only later, while others may partially normalize or be compensated for as the brain develops. Importantly, apparent normalization of one molecular or cellular feature does not necessarily mean that the developmental consequences of the earlier disturbance have disappeared.

For me, the most important unresolved question is therefore: which early changes caused by ASH1L deficiency actually initiate later neural and behavioral dysfunction? Many molecular changes can be detected in a mutant brain, but they are unlikely to be equally important. We need better temporal and cell-type-specific studies to distinguish primary causal defects from downstream consequences and compensatory responses. Identifying those initiating events—and the developmental windows during which they remain modifiable—will be particularly important for thinking about future therapeutic strategies.

Animal and cellular models are essential for understanding ASH1L biology, but no model should automatically be assumed to reproduce the biology of a particular human ASH1L variant. Many experimental models, including some of our own, use strong or tissue-specific loss of Ash1l to reveal its biological functions. These models can establish causality and identify mechanisms, but they are not necessarily equivalent to the constitutional heterozygous variants found in affected individuals.

Before using a model as a mechanistic bridge to a specific human allele, I would first want to establish what that allele actually does. Does it reduce ASH1L RNA or protein, produce a stable altered protein, or impair a particular function such as catalytic activity or chromatin engagement? Ideally, the model should reproduce that primary molecular consequence and relevant downstream abnormalities in the appropriate cell types and developmental periods.

I would question a model if its molecular consequence is fundamentally different from that of the human allele. For example, a homozygous conditional knockout can be extremely useful for defining ASH1L function, but it should not automatically be viewed as a quantitative model of heterozygous human disease.

For this reason, I see different ASH1L models as complementary rather than interchangeable. The strongest conclusions will come when a mechanism can be reproduced across models and, ultimately, connected directly to the molecular effect of the human allele.

I think we need a relatively high threshold before improvement in an ASH1L model should influence human treatment research. Improvement in a single behavioral assay, molecular marker, or cellular phenotype is encouraging, but by itself it does not demonstrate that the underlying disease mechanism has been corrected.

Our published work with vorinostat provides an example. Postnatal treatment of Ash1l-deficient mice improved sociability and object-recognition performance, showing that at least some abnormalities in this model remain modifiable after birth. That was important proof of principle, but it should not be interpreted as evidence that vorinostat is an established treatment for people with ASH1L-related disorders.

For me, a convincing translational result would connect several levels of evidence. The intervention should modify the molecular mechanism that motivated the treatment, improve relevant cellular or circuit abnormalities, and produce reproducible functional improvement. We also need to know whether the benefit persists after treatment ends, whether there are important adverse effects, and whether the effect occurs in disease-relevant heterozygous or allele-specific models.

It is also important to distinguish several very different outcomes. Prevention means intervening before a developmental abnormality emerges. Developmental support may help the nervous system follow a more favorable trajectory without correcting the original molecular defect. Compensation improves function despite persistence of the primary defect. Durable correction implies sustained normalization of an important causal mechanism. And reversal is the strongest claim: it requires showing that an already-established abnormality can be restored after it is present.

These distinctions matter especially for neurodevelopmental disorders because treatment during development can change the trajectory of the system without necessarily reversing everything that occurred earlier. I therefore think the most informative experiments will combine mechanism, developmental timing, durability, and functional outcome rather than relying on any single rescue measurement.