Dr. Suhua Chang

Dr. Suhua Chang is an internationally recognized researcher and professor, affiliated with the Peking University Institute of Mental Health in Beijing, China. Her work plays a major role in advancing our understanding of the complex relationship between genes, cognition, and mental health, offering valuable insights into potential identification, prevention, and treatment strategies for both neurodevelopmental and psychiatric disorders. In her most recent work, she and her team uncovered distinct genetic profiles underlying common ASD phenotypes. We thank you, Dr. Chang, for sharing your insights with the ELePHiNt Community. Enjoy!

Q1:  You analyzed transcriptomic data from over 1,700 individuals with ASD — one of the largest such cohorts ever studied. What motivated you to look for molecular subtypes rather than treating ASD as a single condition?

The motivation, says Dr. Chang, is today’s clinical reality. “Two children who both meet diagnostic criteria for ASD can look extraordinarily different — one may be largely nonverbal with intense repetitive behaviors, another may be highly verbal with relatively subtle social difficulties”. Whether or not these presentations represent different subtypes of ASD, she goes on to note, has been a topic of active discussion for several years now. “The theoretical case for subtyping had already been made on the genetic level”, says Dr. Chang, pointing to pre-existing studies suggesting that common ASD phenotypes have different genetic underpinnings. “What was missing was a data-driven, bottom-up approach that started from molecular data rather than clinical observation”. This study’s key advancement, she goes on to argue, is the size of the dataset. “The availability of RNA-seq data from over 1,700 individuals in the Simons Simplex Collection gave us the sample size to attempt that kind of unsupervised decomposition with reasonable statistical power.” 

Q2. Your team identified three molecular subtypes, each with a distinct clinical signature. Were any of these findings surprising, or did they align with what clinicians have long suspected from the behavioral side?

“In some ways the broad strokes confirmed what clinicians have suspected”, says Dr. Chang. Specifically, the analysis distinguished between individuals whose symptoms were driven primarily by social-communication difficulties and those characterized more strongly by repetitive behaviours, while also identifying a group with relatively mild overall presentation. “Those patterns have been described in behavioral and factor-analytic studies before [...] so finding them emerge independently from gene expression data was reassuring rather than surprising”. Rather than revealing entirely new categories of ASD, then, the findings provide biological evidence for distinctions that clinicians and researchers have long observed. 

Q3. The data came from lymphoblastoid cell lines rather than brain tissue. What were the hardest trade-offs you had to make in study design, and how much do you think the tissue source shapes what you can and cannot see?

The choice of starting material comes with an obvious but necessary limitation, says Dr. Chang. “The reason we used lymphoblastoid cell lines is straightforwardly practical— they gave us access to over 1,700 samples with matched phenotypic data, which is simply not achievable with postmortem brain tissue.” More specifically, downstream analyses such as unsupervised clustering required sample sizes that are not currently achievable with postmortem brain samples. The use of lymphoblastoid cell lines reflects this practical constraint, then, though—as noted by Dr. Chang—prior work suggests that these samples likely also capture many molecular signatures seen in peripheral blood, including those potentially relevant to neural biology. 

“On the other hand” says Dr. Chang, “we admit that brain tissue would be better to identify the pathology. The tissue source almost shapes what we see. Immune-related signals, for instance, may be more prominent in a peripheral immune cell-derived line than they would be in cortical neurons. Conversely, subtle neuronal signals may be attenuated or absent.” To address this concern, the team validated their findings in a subset of post-mortem brain tissue. “[...] the fact that subtype-level pathway patterns—particularly the co-occurrence of upregulated immune pathways and downregulated neuronal pathways in Cluster 1—were reproduced in postmortem cortical tissue gives us more confidence that the signal is not purely an artifact of the tissue type.” Given the size of the validation cohort, however, Dr. Chang also cautions against overinterpreting the results. 

Q4. If a clinician walked out of this talk and wanted to start using these subtypes in their practice tomorrow, what would you tell them they can and cannot do with this information today?

“What a clinician can take from this work, we think, is a conceptual framing” says Dr. Chang. “The evidence that social-communication deficits and restricted repetitive behaviors may have separable biological underpinnings is consistent with what many experienced clinicians have intuited — that these are not just two dimensions of the same thing. That framing might usefully inform how clinicians think about heterogeneity within their patient populations, and how they interpret variable treatment responses.” Importantly, however, as noted by Dr. Chang, while these findings may encourage a shift in clinical perspective, they do not yet provide sufficient evidence to support new diagnostic profiles or therapeutic approaches. “Maybe in future, if you get the transcriptome data of a patient’s peripheral blood, we can classify [them] into one of [the] subtype[s], but this clinical utility [requires] larger replication studies, the development of reliable biomarker-based assignment tools, and ultimately clinical trials that test whether subtype-informed stratification improves outcomes. We are not there yet.”

Q5. Some of your genetic-burden findings were described as preliminary trends that did not reach statistical significance. How do you navigate the tension between wanting to report everything the data hints at and the discipline of sticking to what survives rigorous correction?

For Dr. Chang, the decision came down to a familiar tradeoff between interpretability and completeness, with the team ultimately prioritizing full transparency. “If we had simply omitted the genetic burden analyses because they did not survive FDR correction, a reader would reasonably ask why we did not look at genetics at all, and future researchers trying to build on or replicate this work would not have the full picture. The broader principle we try to follow is: report what you found, be precise about the strength of the evidence, and let the reader calibrate their confidence accordingly.” 

In practice, this shaped how the results were presented: non-significant genetic signals were reported, but kept distinct from the study’s main conclusions. “[...] we were deliberate about how we framed those findings — as preliminary trends that require replication in larger, independent cohorts before we draw reliable conclusions. We did not build interpretive arguments on top of non-significant results, and we did not allow those results to drive our main conclusions. The subtyping itself, the phenotypic characterization, and the transcriptomic pathway analyses are where the main findings lie, and those are what we emphasize.”

Q6. What is the next experiment or study you most want to see done to build on this study's findings?

Dr. Chang’s priority now is testing these findings in larger, more deeply characterized cohorts: “[...] with transcriptomic data from peripheral blood paired with rich phenotypic characterization, [we could obtain] more rigorous validation of our findings.” She also highlights possible clinical application for guiding targeted treatment as a key future direction. “If the subtype can be validated in an independent cohort, the next step is to determine how the subtyping results can be used to guide targeted treatment. For instance, immune-based interventions might be appropriate for the immune dysfunction group, while other groups may benefit from alternative therapies. In the future, well-designed clinical trials will be needed to validate subtype-specific treatment response predictions. ”

Interview conducted by: Dominique Lumley

Blog-post written by: Dominique Lumley