The technical premise rests on normative modeling, a statistical framework that has become one of the most influential tools in modern neuroimaging. Instead of comparing patients with a small control group and asking whether a brain region is, on average, larger or smaller, normative modeling fits reference curves — much like the growth charts pediatricians use for height and weight — that describe the expected range of brain measures at every age. Each individual scan is then assigned a centile, or deviation score, quantifying how far that person’s brain structure sits from the population norm. Applied to thousands of scans, this approach turns a single clinical MRI into a rich phenotypic fingerprint, one that can flag extreme deviations and map where in the brain they concentrate.

What makes the new work distinctive is its source of data. Prospective neuroimaging studies routinely exclude individuals with severe cognitive impairment, significant medical comorbidities, or an inability to tolerate a research MRI session without sedation — features that are common in 22q11.2 deletion syndrome and other rare genetic disorders. Those exclusions quietly narrow the phenotypic spectrum that research cohorts can see, limiting how generalizable the findings are to the full population of affected patients. Clinical scans, by contrast, are acquired for medical reasons and include precisely the patients that research studies tend to leave out. In this retrospective, cross-sectional study, the team analyzed 92 patients with 22q11.2 deletion syndrome, aged six months to 21 years and 49 percent female, alongside 252 matched clinical controls, all drawn from hospital records at the Children’s Hospital of Philadelphia.

The reference curves themselves were derived from 1,995 clinical scans, a large normative sample that allowed the researchers to estimate age-specific expectations for a battery of structural brain measures: total gray matter volume, white matter volume, subcortical gray matter volumes, intracranial volume, cortical thickness, and regional cortical surface area and volume. The statistical machinery behind these charts, generalized additive models for location, scale, and shape, is the same family of models used to construct conventional child growth standards, and it can capture how both the average and the variability of brain measures change across development. Every patient scan was then scored against these population norms, producing deviation maps that could be compared across individuals and aggregated to characterize the syndrome as a whole.

The headline result is a picture of widespread structural reduction. Patients with 22q11.2 deletion syndrome showed broad decreases in brain volumes, with the largest effect reaching a Cohen’s d of −1.31 — a very large effect size by any neuroimaging standard — and reductions in cortical surface area peaking at d = −0.71. Cortical thickness, however, moved in the opposite direction, increasing to d = 0.39. That combination is technically meaningful: reduced surface area with preserved or increased thickness suggests that the cortex in this syndrome is affected in its areal expansion rather than simply being thinned, a dissociation that implicates specific developmental processes in the folding and areal patterning of the cerebral sheet. Subcortical structures and white matter were also reduced, consistent with the syndrome’s known impact on deep gray matter nuclei and connecting pathways.

Because clinical samples invite skepticism about scan quality and processing, the team validated their findings against an entirely independent source: the ENIGMA-22q research consortium, a multinational collaboration that has assembled prospectively collected research MRI data from 22q11.2 deletion syndrome patients across dozens of sites. The spatial convergence was striking. Regional patterns of volume and surface area deviations in the clinical cohort correlated with those in the research cohort at r = 0.61 to 0.87, indicating that the two approaches — one built on routine hospital scans, the other on carefully standardized research acquisitions — were seeing the same underlying biology. This is a critical proof of principle that heterogeneous clinical imaging, processed with appropriate quality controls, can robustly characterize brain structure in rare genetic disorders at a scale and inclusiveness that research cohorts alone cannot match.

Within-syndrome heterogeneity emerged as a central theme. Forty percent of patients showed at least one global brain measure below the 2.5th centile of the population norm — a rate of extreme deviation far exceeding what would be expected by chance — yet individual patients varied enormously in which measures were affected and how severely. Standard growth charts, the authors note, can identify patients who deviate from typical development, but they cannot distinguish a patient whose brain is small because of the syndrome from one who is disproportionately affected relative to their genetic peers. That distinction matters clinically, because the patients who sit at the extreme tail of their own syndrome’s distribution may face different risks and may need different monitoring than those who cluster near the syndrome’s average.

To address that gap, the team generated syndrome-specific growth charts — reference curves built from the 22q11.2 deletion syndrome patients themselves, so that each patient’s brain measures could be scored against the expectations of their genetic peers rather than the general population. In exploratory analyses, deviations in cerebellar volume relative to other 22q11.2 deletion syndrome patients were nominally associated with lower language scores across two independent assessment methods, the Clinical Evaluation of Language Fundamentals and the Preschool Language Scales, with p-values below 0.03. The cerebellum, long appreciated for its role in motor coordination, has increasingly been implicated in language and cognition, and the finding hints that cerebellar development may carry prognostic information about cognitive and linguistic outcomes in this syndrome. The authors are careful to frame these analyses as exploratory, but the framework they introduce — scoring individual brain structure against syndrome-specific norms and testing whether those scores predict outcomes — is a novel route toward risk stratification in rare neurogenetic conditions.

The study also reached back to the genome itself. Spatial patterns of cortical volume and surface area deviation across the cortex correlated with the expression of genes located within the 22q11.2 locus, as mapped by the Allen Human Brain Atlas, a publicly available resource that charts where thousands of genes are expressed across the human brain. This gene-expression alignment suggests that the regional pattern of brain alteration is not random but reflects, at least in part, the spatial distribution of the very genes whose absence defines the syndrome. It is a technically elegant convergence: the deletion removes a set of genes, those genes are expressed unevenly across the cortex, and the cortex deviates unevenly in exactly the regions where the deleted genes are most active.

The implications extend well beyond a single microdeletion. 22q11.2 deletion syndrome affects an estimated one in roughly 2,000 to 4,000 births and is among the most common copy number variants involving the brain, and its study has long served as a model for understanding how CNVs shape neuropsychiatric risk, including schizophrenia, autism spectrum disorders, and attention-deficit/hyperactivity disorder. If routine clinical MRI can be transformed into syndrome-specific brain charts for this condition, the same pipeline — normative modeling from large clinical reference samples, deviation scoring, validation against research consortia, and linkage to cognitive outcomes — could in principle be applied to other rare genetic disorders whose patients are systematically underrepresented in research imaging. The work, funded by the National Institute of Mental Health, was conducted under ethical review at CHOP and Penn, with retrospective use of clinical scans determined exempt because it involved secondary use of existing data with low risk to human subjects. What the study delivers, above all, is a scalable framework: a way to turn the scans hospitals already acquire into quantitative, individualized maps of brain development that can include the very patients research studies have historically left behind.

Subject of Research: Brain structure in 22q11.2 deletion syndrome mapped with clinical neuroimaging and normative modeling

Article Title: Charting brain structure in 22q11.2 deletion syndrome with clinical neuroimaging

Article References: Jung, B., Schmitt, J. E., Seidlitz, J., Schabdach, J. M., Karandikar, S., Crowley, T. B., Dorfschmidt, L., Mandal, A. S., Zimmerman, D., Williams, R. M. S., Prem, S., Levitis, E., Gardner, M., Cyr, K., Padmanabhan, V., Taylor, J. H., Ruparel, K., Boen, R., Bearden, C. E., … O’Hara, R. (2026). Charting brain structure in 22q11.2 deletion syndrome with clinical neuroimaging. Genome Medicine. https://doi.org/10.1186/s13073-026-01788-7

Image Credits: AI Generated

DOI: 10.1186/s13073-026-01788-7

Keywords: 22q11.2 deletion syndrome, normative modeling, clinical neuroimaging, ENIGMA-22q, Genome Medicine, brain growth charts, schizophrenia risk, copy number variant, Charting, brain, structure, deletion

Temporal lobe epilepsy is the most common form of focal epilepsy, and its best-understood cause is hippocampal sclerosis, a characteristic scarring of the brain’s memory hub that is readily detectable on dedicated epilepsy protocols. Yet in up to 30 percent of patients, even high-resolution scans reveal nothing. This so-called MRI-negative temporal lobe epilepsy, abbreviated TLE-MRIneg, represents a genuine clinical crisis. Without an identifiable epileptogenic lesion, many patients are deemed unsuitable for curative surgery, leaving them dependent on anti-seizure medications that often fail to control their seizures. The new multicenter study, conducted under Italy’s national “3TLE Project” (NET-2013-02355313), argues that the field has been treating this group as an undifferentiated whole when it should instead be parsed into biologically meaningful subgroups.

A central target of the investigation was radiological amygdala enlargement, a subtle finding that has increasingly been proposed as a biomarker in MRI-negative cases. Enlargement of the almond-shaped amygdala, a structure deep in the temporal lobe central to emotion and seizure generation, has been reported in anywhere from 12 to 63 percent of MRI-negative patients, depending on the study. The researchers used volumetric analysis of three-dimensional T1-weighted and fluid-attenuated inversion recovery (3D-FLAIR) sequences to quantify amygdala size, normalizing measurements to intracranial volume to guard against the confounding effects of overall head size. Their goal was to determine whether amygdala enlargement truly marks a distinct disease entity or is merely an incidental observation with questionable epileptogenic significance.

The cohort itself was assembled prospectively between 2016 and 2020 and ultimately comprised 172 patients with temporal lobe epilepsy, 96 classified as MRI-negative and 76 as having hippocampal sclerosis. Diagnosis followed International League Against Epilepsy guidelines and was established by board-certified neurologists using neurological examination, clinical history, routine EEG, and video-EEG monitoring where available. Side of seizure onset, left, right, or bilateral, was assigned through consensus among clinicians who integrated interictal and ictal EEG findings with seizure semiology. Crucially, the MRI-negative patients were further subdivided according to whether their seizures appeared to arise from the mesial temporal structures or from the lateral neocortical temporal surface, a distinction drawn from previously published electrophysiological evidence.

The analytical strategy was deliberately multimodal. Rather than testing a handful of preselected hypotheses, the team deployed a battery of multivariate techniques, including principal component analysis, partial least squares, and multivariate analysis of covariance, to extract patterns linking imaging measures, demographic variables, and clinical features. Clustering quality was assessed with the Dunn Index, a metric that quantifies how well-separated the resulting groups are, and group-level statistical contrasts were corrected for multiple comparisons using the false discovery rate procedure. Nearest neighbor matching was applied to reduce confounding when comparing MRI-negative patients with their hippocampal sclerosis counterparts, a design choice that strengthens causal inference in observational clinical data.

The results delivered a pointed message: MRI-negative temporal lobe epilepsy and hippocampal-sclerosis-associated epilepsy behave as distinct entities, not points on a shared pathophysiological continuum. The clinical tables tell part of the story. Age of seizure onset in the MRI-negative group averaged 34.19 years, compared with just 18.72 years in the hippocampal sclerosis group, a difference of more than fifteen years that far exceeded what chance would allow. Mean age in both groups was similar, around 42 to 43 years, and the sex distribution was nearly balanced, underscoring that the divergence between the two syndromes lies specifically in the timing and character of disease emergence, not in who happens to be affected.

That age-of-onset gap is more than a statistical curiosity. Early-onset temporal lobe epilepsy, as seen in hippocampal sclerosis, is typically linked to childhood febrile seizures and a progressive sclerotic process that carves a visible lesion into the hippocampus over years. A substantially later onset, as documented in the MRI-negative group, points to different mechanisms altogether, possibly involving subtle developmental anomalies, inflammatory changes, or amygdala-centered circuit dysfunction that conventional magnetic resonance resolution cannot capture. The authors argue that lumping these patients together in prior research, based on the hypothesis of a continuum, has obscured precisely the heterogeneity that individualized treatment now demands.

The implications for epilepsy surgery are among the most consequential. Curative resective surgery for temporal lobe epilepsy is most successful when a concordant lesion can be identified, and patients with hippocampal sclerosis enjoy among the best surgical outcomes of any epilepsy syndrome. Patients with MRI-negative disease have historically been triaged away from the operating room, or subjected to more invasive and riskier evaluation, because the epileptogenic zone could not be localized. By demonstrating that MRI-negative cases carry identifiable clinical signatures, such as later onset and probable mesial or lateral onset patterns, and potentially measurable imaging correlates such as amygdala enlargement, the study opens a path toward rational candidate selection for surgery even in the absence of a visible lesion.

The methodological framework itself may prove as influential as the clinical findings. By combining volumetric region-of-interest analysis with data-driven multivariate decomposition, the researchers sidestepped the trap of looking only for what they expected to find. Principal component analysis and partial least squares allow latent variables, hidden axes of variation that combine imaging and clinical information, to surface from the data organically. This approach mirrors the growing role of artificial intelligence in neuroimaging, where machine learning models detect cortical and subcortical biomarkers invisible to the human eye, but it retains an interpretability advantage: each latent variable can be traced back to the specific clinical and volumetric measures that compose it.

The research team frames its conclusions as a corrective to a decade of conflation. Previous studies frequently pooled MRI-negative and hippocampal sclerosis patients, implicitly treating the former as a milder or earlier-stage version of the latter. The new evidence, building on earlier clinical investigations by the same group, dismantles that assumption. If the two syndromes differ in onset age, semiology, neurophysiology, and imaging profile, then research findings derived from mixed cohorts may have systematically underestimated differences and muddied biomarker discovery. Separating them, the authors contend, is a prerequisite for the development of targeted research strategies and genuinely personalized epilepsy medicine.

The stakes extend beyond the clinic into basic neuroscience. The amygdala, and in particular its basolateral nucleus, has emerged as a candidate epicenter for a subset of MRI-negative cases. Amygdala enlargement, if validated prospectively as an epileptogenic marker rather than a bystander phenomenon, could redefine how these patients are counseled, monitored, and treated. It could also inform the development of neuromodulation targets, since stimulation and ablative therapies increasingly rely on precise anatomical hypotheses about where seizures begin. The heterogeneity documented in this study suggests that a one-size-fits-all approach to MRI-negative disease would fail precisely because the underlying circuits differ from patient to patient.

What the study does not yet settle is causation. Whether amygdala enlargement is a cause, a consequence, or a correlate of seizure generation in MRI-negative patients remains an open question, and the authors are explicit that its significance is still contested in the literature. Longitudinal follow-up, higher-field imaging, and histopathological confirmation from surgical specimens will be needed to anchor the phenotypes they describe in tissue-level biology. Still, the multicenter design, prospective recruitment, and statistically disciplined analysis give the findings unusual weight for a condition defined, until now, largely by what scans fail to show.

For the estimated hundreds of thousands of patients worldwide whose temporal lobe epilepsy resists conventional imaging, the message is one of cautious hope. The label “MRI-negative” has long functioned as a diagnostic dead end, a shrug in radiological form. This study transforms that label into a starting point for subtyping, prognosis, and therapy selection. In an era when epilepsy care is inching toward the precision model that transformed oncology, the demonstration that invisible epilepsy has visible structure, detectable in onset patterns, electrical signatures, and volumetric subtleties, may mark the moment the field stopped searching for a single answer and started asking the right questions.