Why the terms differ. Global developmental delay is a provisional term for children under ~5 years, when IQ testing is unreliable; it deliberately acknowledges uncertainty, because a young child's trajectory is not yet fixed — some catch up, others meet criteria for intellectual disability once formal cognitive and adaptive testing becomes valid. ASD is defined behaviorally (social-communication deficits plus restricted/repetitive behavior; ~1 in 36 children), and intellectual disability co-occurs in a substantial minority — the two increasingly look like overlapping, partly shared genetic architecture rather than separate silos.

The shift, and why it happened. A generation ago, genetic testing followed syndrome recognition — you tested when a child 'looked like' a known condition. Three things changed that: the recognition that de novo variants drive much of severe NDD, the arrival of affordable trio sequencing, and professional guidelines (ACMG 2021) endorsing exome/genome as a first- or second-tier test. The clinician's job shifted from matching a gestalt to ordering and interpreting a structured work-up.

The de novo paradox. Families often reason that an unremarkable family history makes a genetic cause unlikely — for severe NDD the opposite holds. Strongly deleterious, early-onset variants are under heavy negative selection (affected individuals rarely reproduce), so the responsible variants are continually re-created de novo rather than inherited. That is exactly why most severe NDD is sporadic, and why 'nothing runs in our family' should not lower suspicion. Even isolated, nonsyndromic delay carries meaningful yield.

The genetic causes of GDD/ID/ASD are strikingly heterogeneous, and the categories are worth understanding because each maps to a different test.

De novo dominant variants dominate severe, sporadic NDD — enriched here for an evolutionary reason: strongly deleterious variants are removed from the population each generation, so they recur as new mutations. Practically, this is why trio sequencing (proband + both parents) is so powerful — comparing the child to both parents immediately flags variants new in the child, roughly doubling yield versus a singleton (OR ~2.04; Clark et al. 2018) and sparing rounds of follow-up parental testing. The deeper reason these critical genes hold so little inherited variation is a survivorship-bias effect: population databases such as gnomAD capture mostly the variation compatible with survival and reproduction — much as the WWII bombers that returned were hit everywhere except the engine. Loss of function in a strongly constrained neurodevelopmental gene is removed by selection rather than transmitted, so the causal variants instead arise de novo (see gene constraint & survivorship bias).

Copy number variants are a parallel, non-overlapping layer. Chromosomal microarray detects ~10% of cases and finds recurrent genomic disorders (16p11.2, 22q11.2, 15q11.2, 1q21.1, and others) that arise at meiotic recombination hotspots — variants short-read exome calling can miss. CMA and sequencing detect different classes of variant and are complementary, not redundant.

Monogenic causes number in the hundreds of genes, many converging on a few pathways (chromatin remodeling, transcription, synaptic function). Yield depends on test and phenotype: trio exome reaches ~30–45% in ID, and for GDD the yield can exceed >50% with trio genome sequencing — or with trio exome paired with CMA for CNV calling (exome alone calls CNVs poorly).

X-linked ID explains part of the long-noted male excess in ID; FMR1 (Fragile X) is the single most common, but dozens of other X-linked genes contribute.

Yield tracks the phenotype through selection and ascertainment: syndromic, severe, early-onset presentations are enriched for Mendelian causes and yield highest, while isolated ASD without ID sits lowest (WES ~10–15%, rising to ~25–30% with comorbid ID). See Diagnostic Yields and CNV Interpretation.

A genetics-oriented exam is not a hunt for minor dysmorphism — most isolated minor features (e.g., epicanthal folds) carry little diagnostic weight. The goal is the handful of findings that genuinely change the differential or the test you order.

Head circumference is informative because it is a readout of brain growth, and the two directions point to different biology. Macrocephaly with DD/ASD suggests overgrowth signalling (the PI3K-AKT-MTOR pathway; PTEN) — PTEN testing is specifically indicated in ASD with macrocephaly because of its tumour-surveillance implications — and several CNVs (notably 16p11.2 deletion). Microcephaly instead points toward genes governing neuronal proliferation (primary microcephaly/MCPH genes such as ASPM; DYRK1A); critically, congenital microcephaly implies a primary growth problem, whereas acquired/progressive microcephaly signals a regressive process such as Rett.

Hypotonia is common and non-specific but still useful: central hypotonia broadens the differential to Prader-Willi, Angelman, maternally-inherited myotonic dystrophy, congenital myopathies/dystrophies, and many CNV syndromes — and helps decide whether to add neuromuscular testing.

Active regression is the finding that should change your tempo. Loss of previously-acquired skills is a red flag for a metabolic or neurodegenerative process — some of which are treatable and time-sensitive — and warrants an expedited work-up: rapid exome/genome, brain MRI and EEG, and targeted metabolic studies, rather than the routine outpatient pace.

Growth abnormality, organomegaly, and congenital anomalies round out the high-yield findings — anomalies in particular raise CMA and exome yield.

A recurring misconception is that genetic NDDs announce themselves with a striking facial gestalt. Most do not — they are recognizable syndromes with subtle, easily-missed features, which is exactly why they are usually diagnosed molecularly rather than by sight. Two implications follow: keep a low threshold for testing, and learn a few high-value clues rather than trying to memorize faces.

Many of these genes cluster in a few functional pathways, which is part of why they share features. Chromatin and transcriptional regulators are heavily represented — ARID1B (Coffin-Siris; hypoplastic fifth finger/toenail, coarse features), SETD5, KAT6A, and EHMT1 (Kleefstra; synophrys, hypotonia) — alongside synaptic and signalling genes such as SHANK3 (Phelan-McDermid; neonatal hypotonia, absent speech, large fleshy hands) and ADNP (Helsmoortel-Van der Aa, a frequent single-gene ASD cause; early primary tooth eruption). Others carry their own clue: DYRK1A (microcephaly, deep-set eyes), KANSL1 (Koolen-de Vries; an amiable disposition and epilepsy), MED13L, RAI1 (Smith-Magenis; inverted-melatonin sleep disturbance and self-injury), and SON (ZTTK; distinctive face with brain MRI abnormalities).

Overgrowth and CNV causes sit alongside these. PTEN and MTOR link to macrocephaly (above), and recurrent CNVs — led by 16p11.2 — are a major, CMA-detectable category, many showing reciprocal deletion/duplication phenotypes (16p11.2 deletion tends toward macrocephaly, the duplication toward microcephaly).

The unifying lesson: subtle phenotypes mean an unremarkable exam does not exclude a genetic cause — the diagnosis comes from the test. (A newer cause, RNU4-2, is also a recognizable phenotype but sits in a non-coding gene — see Testing Strategy.)

From CMA-first to sequencing-first. The historical first-tier — CMA plus Fragile X — has not disappeared, but exome/genome has moved forward because trio sequencing has the highest single-test yield. The trade-off is interpretive load: broad sequencing produces variants of uncertain significance and can surface secondary findings, so it is best paired with pretest genetic counseling (ideally a certified genetic counselor) to set expectations and obtain meaningful consent. Where counseling access is limited, starting with CMA + Fragile X is reasonable. Crucially, exome does not make CMA and Fragile X obsolete — standard exome reliably detects neither CNVs nor the FMR1 repeat — so they remain part of the work-up.

A reasonable default: CMA + Fragile X in essentially all unexplained GDD/ID, trio exome or genome as first- or second-tier (first-line when counseling is available and the phenotype is severe/syndromic), and targeted testing when the exam points somewhere specific.

A negative result is a checkpoint, not an endpoint. Gene discovery is rapid, so reanalysis of an older exome adds yield over time. And some causes are simply not in the coding exome: RNU4-2 (ReNU syndrome) — a non-coding U4 small-nuclear-RNA gene — produces a recognizable phenotype yet is now estimated at ~0.4% of all NDD, among the most common monogenic causes, and was invisible to coding-focused pipelines (Chen et al., Nature 2024; RNU2-2 is similar). This is a concrete argument for genome sequencing and for revisiting unsolved cases.

Counsel recurrence by mechanism. A de novo variant carries a low but non-zero recurrence (~1%, from parental gonadal mosaicism); inherited recessive, X-linked, or familial-CNV causes carry higher, pattern-specific risks. Beyond recurrence, a precise diagnosis enables syndrome-specific surveillance, sharper prognosis, occasional precision treatment, and an end to the diagnostic odyssey. See Genetic Counseling and Diagnostic Yields.