A high-quality study of subtypes of autism is available

What was this study and what does it tell us about teaching our kids?

On 8 July 2025, Aviya Litman, Natalie Sauerwald, Olga Troyanskaya,¹ and their colleagues of Princeton University published a report of their research about the genetic basis of autism in Nature Genetics. In contrast to simple genetic explanations that search for an individual gene or mutation that underlies autism, the Princeton team took a broader and sophisticated approach. They scoured data from many sources to look for patterns of behavioral manifestations of autism and then match those patterns with genetic variations.

In this article I hope to provide guidance for readers of Special Education Today about the study itself, some links for media coverage of it, some context surrounding it, and some thoughts about its implications for the practice of special education. This ain’t not no full-on [sic] rundown on subtypes of autism, but I hope it is informative.

The study

Please start with the abstract from the Littman report—and don’t give up the ship as you read it, because you might find it a bit opaque (it is a challenge):

Unraveling the phenotypic and genetic complexity of autism is extremely challenging yet critical for understanding the biology, inheritance, trajectory and clinical manifestations of the many forms of the condition. Using a generative mixture modeling approach, we leverage broad phenotypic data from a large cohort with matched genetics to identify robust, clinically relevant classes of autism and their patterns of core, associated and co-occurring traits, which we further validate and replicate in an independent cohort. We demonstrate that phenotypic and clinical outcomes correspond to genetic and molecular programs of common, de novo and inherited variation and further characterize distinct pathways disrupted by the sets of mutations in each class. Remarkably, we discover that class-specific differences in the developmental timing of affected genes align with clinical outcome differences. These analyses demonstrate the phenotypic complexity of children with autism, identify genetic programs underlying their heterogeneity, and suggest specific biological dysregulation patterns and mechanistic hypotheses.

Most readers of Special Education Today are well aware of the idea that the individuals with autism are not a homogenous group. Two very different individuals might both be “on the spectrum.” A young child with little or no communication skills, difficulty interacting with others, stereotypical behaviors, and developmental delays might be identified as having autism and an adult who is loquacious, had few or no developmental delays, and might have some hyperactive behavior as well as a strong interest in orderliness might also be identified.

One can generate many different case descriptions. Individuals with autism represent such a diverse group that learning about an individual case provides little general knowledge about autism. Readers may have heard the saying, “if you know one child with autism, you know one child with autism.”

The range of variation among cases provides a big hint that there might be clumps, clusters, or subtypes of autism. For their study, the Princeton team of Professor Troyanskaya went well beyond studying an individual case. Using the SPARK² database, they assembled data about more than 5300 individuals. The drew together almost 240 bits of data about each of those individuals from standard diagnostic instruments such as the Child Behavior Checklist (Achenbach & Rescorla, 2000). The data covered a wide range of characteristics related to autism. The researchers aggregated the phenotypic data into measures of (a) limited social communication, (b) restricted or repetitive behavior, (c) attention deficit, (d) disruptive behavior, (e) anxiety or mood symptoms, (f) developmental delay, and (g) self-injurious behavior.

So, for each of the > 5000 individuals in their study, they had a profile composed of the seven measures. They analyzed all these profiles to find patterns of higher and lower scores. They found four subtypes of individual patterns,³ which they characterized as follows:

1. A group they called “social-behavioral” (n = 1976) with difficulty across social communication, repetitive behavior, disruptive behavior and attention deficit, but without developmental delays;

2. A group they called “mixed ASD with DD” (n = 1002) who showed a mixture of higher and lower problems with repetitive behavior, social communication, and higher levels of developmental delay and self-injury;

3. A group they called “moderate challenges” (n = 1860) who scored lower than others on almost all types of phenotypic measures; and

4. A group they called “broadly affected” (n = 554) who scored higher than others on almost all the measures.

The accompanying image shows the profiles for these groups. The groups represented are “moderate challenges” as gold, “broadly affected” as red, “social behavioral” as green, and “mixed ASD with DD” as blue.

The Princeton team went much farther than simply describing the clusters. They found that the clusters aligned with both parents’ reports about problems, reports that were not used in the statistical analyses to create the clusters, and clinicians reports of co-morbidities. Such data bolster the assertion that the clusters are not just statistical artifacts, that the clusters the researchers identified have connections to the reality of autism.

They also found that when the applied the same algorithm to an independent set of descriptions of cases, they got similar results. That is to say, the cluster solution replicated. They seemed to have found clusters that had corresponded with other findings about individuals with autism; there was external validity.

A critically important finding, however, wasn’t about the internal validity of the clusters or subtypes. Nor was it about the fit with other evidence about the characteristics of individuals with autism (i.e., some external validity). It was a special case of external validity: When they matched their findings to genetic findings about other disorders there were important connections. They were, for example, able to point to de novo mutations for the “broadly affected” group. The social-behavioral class seemed to have smaller genetic connections. Other known genetic factors seemed to be separately connected to the “mixed ASD and DD” group. And there is a shipload of additional findings that this study reports.

These are fascinating advances in understanding the nature and causes of autism spectrum disorders. The research reported by this team—and likely their future research as well as other teams’ research—will help us to have a much more firm conception of autism.

Implications for practice

In the final paragraph of their main report, Litman et al. (2025) state quite simply that “Future research could also examine how interventions may differ among the classes.” Let that sink in to our thinking for a moment or two.

There is nothing in this research that tells special educators, behavior analysts, speech-language clinicians, or others concerned with providing services what to do Monday morning. This is descriptive research, not proscriptive research.

Now, to be sure, future research could examine whether individuals with Type Social-Behavioral Autism might benefit more from Method X instruction but individuals with Type Mixed ASD and DD Autism might benefit from Method M instruction. But this study does not tell us that. Such interpretations are, well, pure speculation.

If special educators and our sisters and brothers are to be able to use this research in our day-to-day work, it is just that we can now say that not only are individuals with disabilities unique individuals, but they are sorta-kinda or four of five sorts of different. That’s good for understanding the disability. As for addressing the problems these individuals experience, we need to continue our efforts to find them early and provide individualized treatment programs that match their unique developmental and educational needs.

We still need to treat individuals with autism as—well—individual cases. Identify their unique educational needs, plan how to address those needs, determine how to ascertain if the planned services are meeting those needs…and then do it all over again in a year or so.

Context

Researchers have worked on the problem of subtypes of autism since the previous millennium. One approach has been less about differentiating clusters and, instead, seeking communalities; the effort by the American Psychiatric Association to incorporate Asperger’s Syndrome, Rett’s Disorder, and other disorders into ASD in might represent this approach. Readers could also remember that the Diagnostic and Statistical Manual once included a category of “pervasive developmental disorder not otherwise specified,” too. The changees in the diagnostic categories from DSM-IV to DSM-5 illustrate clumping. [FN: Dear Readers may also recognize that this clumping is likely to account for some of the growth in the number of individuals who are identified as having ASD.]

Oftentimes, the discussions of subtypes have featured a particular perspective on autism such as play behavior (e.g., Kilinger & Renner, 2000), theory of mind (Prior et al., 1998) or optometry (Jensen et al., 2014). By not beginning with a theoretical bias and starting with specifying a wide array of descriptors and finding clusters among those descriptors, the research by Litman et al. (2025) provided a broader perspective on subtypes of ASD. To be sure, many of those descriptors are not objectively behavioral (they’re reports by caregivers or others), but the atheoretical base provided a welcome conceptual advance.

Media

In addition to reading the actual report of the research, Dear Readers may want to consult non-technical publications that describe it. I see that mass media outlets are reporting the story at about the time of this post. The study is “going viral,” so we are likely to encounter colleagues who want to discuss it.

Here are a few examples:

• Molly Sharlach of Princeton’s engineering communications office wrote about it 9 July 2025 in “Major autism study uncovers biologically distinct subtypes, paving the way for precision diagnosis and care.”

• Ariana Eunjung Cha of the Washington Post published an article entitled “New science points to 4 distinct types of autism: Scientists are redefining autism as a complex condition with multiple forms, challenging traditional notions.”

• Shari Wiseman wrote “Big data for classifying ASD” published 4 August 2025 in Nature Neuroscience.

Readers can find other media examples simply by searching with their favorite search engine. ⁴

References

Achenbach, T. M., & Rescorla, L. (2000). Child behavior checklist. Burlington.

Jensen, K. A., Hoppe, E., Remick-Waltman, K., Spors, F., & Egan, D. (2014). Article 4 Update on Autism Spectrum Disorders for Optometry: A Review of the Literature. Optometry & Visual Performance, 2(5), 220-234.

Klinger, L. G., & Renner, P. (2000) Performance-based measures in autism: Implications for diagnosis, early detection, and identification of cognitive profiles. Journal of Clinical Child Psychology, 29(4), 479-492, https://doi.org/10.1207/S15374424JCCP2904_3

Litman, A., Sauerwald, N., Green Snyder, L., Foss-Feig, J., Park, C. Y., Hao, Y., Dinstein, I., Theesfeld, C. L., & Troyanskaya, O. G. (2025). Decomposition of phenotypic heterogeneity in autism reveals underlying genetic programs. Nature Genetics, 57(7), 1611-1619. https://doi.org/10.1038/s41588-025-02224-z

Prior, M., Eisenmajer, R., Leekam, S., Wing, L., Gould, J., Ong, B., & Dowe, D. (1998). Are there subgroups within the autistic spectrum? A cluster analysis of a group of children with autistic spectrum disorders. The Journal of Child Psychology and Psychiatry and Allied Disciplines, 39(6), 893-902. https://doi.org/10.1111/1469-7610.00389

Tanguay, P. E. (2000). Pervasive developmental disorders: A 10-year review. Journal of the American Academy of Child & Adolescent Psychiatry, 39(9), 1079-1095. https://doi.org/10.1097/00004583-200009000-00007

Footnotes

1. Olga G. Troyanskaya leads the “lab” at the Sigler Institute for Integrative Genomics at Princeton where the team conducting this study worked. She is an accomplished scholar, as reflected in her ORCID profile.

2. SPARK stands for Simons Foundation Powering Autism Research for Knowledge. It is a US national study using a cohort of individuals and families affected by ASD. Individuals with autism and members their families voluntarily enrolled in the project. The project gathers extensive data about individuals and their family members, including medical records, saliva (for genetic analysis), and much more. It is especially valuable because the data collection is well-conceived and -executed and the number of participants is quite substantial.

3. The researchers tested other grouping ranging from just 2 patterns to 10 patterns and used statistical analyses to determine that four was the optimal number. The four-class solution fit the data most accurately, permitted clear interpretation, and yielded stable and robust clumps.

4. I recommend avoiding the “Big Internet” engines. Try DuckDuckGo, Startpage, and other private means of searching (see Surfshark). They don’t spy on your searches and sell the data about what you are seeking or seeing