In a “glass-half-empty” appraisal of the status of autism research, Prof. Laurent Mottron (2021) laments the problem of unresolved heterogeneity plaguing the discovery frontier. Although some of the elements of heterogeneity he cites are undoubtedly problematic (e.g., autism with versus without intellectual disability), his proposal for a wholesale redefinition of the condition by decomposing populations of ASD-affected patients into compartments with homogeneous values for DSM 5 specifiers (i.e., groups of phenocopies) is unlikely to succeed because it does not explicitly account for what breeds true in autism.

Let us take stock of the facts: (a) Autism, even when defined according to less-than-perfect parameterizations of the condition, is highly heritable, on the order of 85% in the general population (Sandin et al., 2017), and most affected children are born to unaffected parents; (b) Subclinical variations of the characterizing traits and features of autism aggregate in the unaffected family members of affected individuals (Robinson et al., 2011), and such variations are continuously distributed population wide (Wagner et al., 2019); (c) Within families in which autism recurs, the individual symptom profiles of co-affected family members paradoxically diverge—this has been known for two decades (Spiker et al., 2002), but its profound implications forgotten, only to be reinvigorated by “rediscovery” of the low twin-cotwin correlations for careful symptom measurements among affected monozygotic twin pairs (Castelbaum et al., 2020; Mazefsky et al., 2008) which Dr. Mottron appropriately points out; (d) The many disparate single-gene conditions that lead to the convergent syndrome of autism are (i) almost never inherited (rather de novo); and (ii) almost always associated with intellectual disability (Myers et al., 2020); (e) What predicts the recurrence of autism in families are joint elevations of heritable early developmental liabilities—some non-specific to autism—which can occur in different combinations or permutations in individual patients (Constantino, 2019; Constantino et al., 2017), and which appear independently heritable (Pohl et al., 2019) and, themselves, continuously distributed in the general population.

What to make of these five observations? First, if any “re-compartmentalization” of autism is to occur, it should separate familial autism from non-familial autism (this is not one of Mottron's proposed specifiers for an autism “prototype”), since inheritance is what CAUSES the vast majority of cases in the population. Notwithstanding the value of sporadic monogenic syndromes in elucidating biology, sibling study designs provide the acid test for what accounts for recurrence of an inherited condition. Second, wading through the epiphenomena of symptomatology after autism develops is not necessarily the way to identify meaningful markers of heterogeneity, because symptom profiles in affected individuals do not breed true. This would be like compartmentalizing hypertension phenocopies on the basis of blood pressure level and the presence or absence of a history of stroke (see below). The paradox is that autism heterogeneity may most reliably be parsed (“fractionated” as it were) before it occurs, not after, and only in accordance with the inherited liabilities that led to it; traces of which may be partially or completely lost or distorted after the condition itself emerges. The current (and finite!) slate of predictors include variation in social reciprocity, attention, hyperactivity, social visual orientation, motor coordination, tactile sensitivity, and cerebellar-dependent learning (Constantino et al., 2021). If this is where the heterogeneity lies, these are promising developmental parameters by which to classify individual cases.

Success in understanding and treating hypertension is instructive here. Because vascular physiology is more easy to study than brain circuitry, Arthur Guyton was able to summarize (in 1972) that variation in the heritable, continuously distributed trait of blood pressure was resolvable to a sub structure of inherited contributors: disparate combinations of variation in vascular resistance, stroke volume, and electrolyte balance could all result in exact phenocopies of elevated blood pressure (Figure 1). A “lucky” feature of hypertension is that those physiologic contributors persist and are measurable in defining different pathways to hypertension, even after hypertension develops, but they are in no way directly reflected by the severity of blood pressure elevation or the comorbidities with which hypertension is associated. In autism, the DSM5 symptoms of the condition—pathognomonic though they may be—are proving themselves to be poor reflections of the combinations and permutations of inherited liability that engender the condition that (secondarily) gives rise to them.

So this is no time to constrain autism to collections of arbitrary-defined or expert-defined phenocopies. We have been through that already. Parse heterogeneity? Yes. Do so by compartmentalization without respect to inheritance? No. Do so on the basis of symptom profiles of affected patients? No, unless consummately aligned with measurement of persistent variation that relates to developmental causes of the condition (eg. ADHD, developmental coordination disorder traits). Study sibling pairs? As often as possible. Identify disparate pathways to the condition? Yes, however I do not believe the hard road ahead is through yet-another novel slicing and dicing of symptom profiles in affected patients. It is more likely to be through developmental research attuned to the indices that predict familial recurrence. And if it is true that those indices are extremes of normally distributed traits in the general population, then the Mottron proposal has missed the point of dimensionality, and a research paradox is that genetic epidemiologic samples will be more informative to new discovery than even the most prototypic of clinical ascertainments. If the common genetic causes of autism are polygenic and incremental, they have been evolutionarily retained in the population for a reason; they are much better understood when studied across the wide range of variation in nature than at the tail of the distribution. In ways that are reminiscent of the Heisenberg Uncertainty Principle, the closer you get to the tail, the more the view gets obscured—for most neuropsychiatric disorders this is likely explained by increased vulnerability to stochastic influences in clinical versus typical populations (Castelbaum et al., 2020; White, 2019), and by the developmental consequences of impairment.

在对自闭症研究现状的“半杯半空”评估中，Laurent Mottron 教授（2021 年）对困扰着研究前沿的未解决的异质性问题表示遗憾。尽管他引用的一些异质性元素无疑是有问题的（例如，有智力障碍的孤独症与没有智力障碍的孤独症），但他建议通过将受 ASD 影响的患者群体分解为 DSM 5 说明符具有同质值的隔间来重新定义这种状况（即表型组）不太可能成功，因为它没有明确说明自闭症中什么是真实的。

让我们来盘点一下事实：(a) 自闭症，即使根据不完美的条件参数化来定义，也是高度遗传的，在一般人群中大约为 85%（Sandin 等人，  2017 年） ，并且大多数受影响的孩子出生于未受影响的父母；(b) 在受影响个体的未受影响的家庭成员中，自闭症的特征性状和特征聚集的亚临床变异（Robinson 等人，  2011 年），并且这种变异在人群中持续分布（Wagner 等人，  2019 年）；（c）在家庭中，自闭症复发，共同受影响的家庭成员矛盾的个体症状型材发散-这已经知道二十年（斯派克等人，  2002年)，但它的深远影响被遗忘了，只有通过“重新发现”低双胞胎相关性才能重新焕发活力，以便对受影响的单卵双胞胎进行仔细的症状测量（Castelbaum 等人，  2020 年；Mazefsky 等人，2008 年）。  Mottron 恰当地指出；(d)导致自闭症收敛综合征的许多不同的单基因病症 (i) 几乎从未遗传（而是从头遗传）；(ii) 几乎总是与智力障碍有关（Myers 等人，  2020 年）；(e) 预测自闭症在家庭中复发的因素是可遗传的早期发育缺陷的共同升高——一些非自闭症特有的——这可以在个体患者中以不同的组合或排列发生（Constantino，  2019 年；Constantino 等人，  2017 年），并且它们似乎可以独立遗传（Pohl 等人，  2019 年），并且它们本身在一般人群中持续分布。

如何看待这五个观察结果？首先，如果要发生任何自闭症的“重新划分”，它应该将家族性自闭症与非家族性自闭症分开（这不是 Mottron 提议的自闭症“原型”规范之一），因为遗传是导致绝大多数的原因人口中的病例数。尽管散发性单基因综合征在阐明生物学方面具有价值，但同胞研究设计为遗传病复发的原因提供了酸性测试。二、自闭症发展后的症状学副现象不一定是识别有意义的异质性标志物的方法，因为受影响个体的症状谱并不真实。这就像根据血压水平和中风病史的存在或不存在划分高血压表型一样（见下文）。悖论是自闭症异质性可能最可靠地在它发生之前而不是之后被解析（“分割”），并且只能根据导致它的遗传责任；在条件本身出现后，其痕迹可能会部分或完全丢失或扭曲。当前（并且是有限的！）预测因素包括社会互惠、注意力、多动、社会视觉取向、运动协调、触觉敏感性和小脑依赖性学习的变化（Constantino 等人， 2021 年）。如果这就是异质性所在，那么这些是对个别案例进行分类的有希望的发展参数。

理解和治疗高血压的成功在这里具有指导意义。由于血管生理学比大脑回路更容易研究，亚瑟·盖顿（Arthur Guyton）能够（在 1972 年）总结说，血压的可遗传的、连续分布的特征的变化可以解析为一个子结构遗传因素：血管阻力、每搏输出量和电解质平衡变化的不同组合都可能导致血压升高的确切表型（图 1）。高血压的一个“幸运”特征是，即使在高血压发展之后，这些生理因素仍然存在，并且可以在定义高血压的不同途径中进行测量，但它们绝不会直接反映在血压升高的严重程度或高血压的合并症上。联系。在自闭症中，该病症的 DSM5 症状（尽管它们可能具有特异性）被证明是遗传责任的组合和排列的不良反映，这些组合和排列导致（次要）引起它们的条件。

因此，现在不是将自闭症限制在任意定义或专家定义的表型集合的时候。我们已经经历过了。解析异质性？是的。在不考虑继承的情况下通过划分来做到这一点？否。是否基于受影响患者的症状特征？否，除非完全符合与疾病发育原因相关的持续变异的测量（例如，多动症、发育性协调障碍特征）。研究兄弟姐妹对？尽可能经常地。确定疾病的不同途径？是的，但是我不相信前方的艰难道路是通过另一种新颖的对受影响患者的症状特征进行切片和切块。更有可能是通过与预测家族性复发的指标相适应的发育研究。如果这些指数确实是一般人群中正态分布特征的极端，那么 Mottron 的提议就错过了维数点，一个研究悖论是，遗传流行病学样本对新发现的信息量甚至比最临床确定的原型。如果自闭症的常见遗传原因是多基因的和渐进的，它们在进化上保留在人群中是有原因的；当在自然界中的广泛变化范围内进行研究时，比在分布的尾部进行研究时，它们的理解要好得多。以让人联想到海森堡测不准原理的方式，越接近尾部， 2020 年；怀特，  2019 年），以及受损的发育后果。