We welcome the central distinction offered in the target article between a latent-variable (i.e., disorder based) and a network theoretical (i.e., symptom based) approach to epidemiology. We have also argued (e.g., Karmiloff-Smith Reference Karmiloff-Smith1992; Reference Karmiloff-Smith1997; Reference Karmiloff-Smith1998; Reference Karmiloff-Smith2007; Reference Karmiloff-Smith2009; Thomas & Karmiloff-Smith Reference Thomas and Karmiloff-Smith2002) in favour of approaches to typical and atypical development that emphasise multi-directional interactions between genes, brain, cognition, behaviour, and environment. Network theory offers precisely such an approach. We also agree with the authors that it is valuable as a way of offering potentially “truer” descriptions according to which diseases are defined as patterns of covariance of symptoms.

In their conclusion, Cramer et al. briefly discuss the potential usefulness of dynamic approaches to network modelling as a way of describing features such as bistable depressive states (Van der Maas et al. Reference Van der Maas, Dolan, Grasman, Wicherts, Huizenga and Raijmakers2006; see also Rolls et al. Reference Rolls, Loh, Deco and Winterer2008) It is disappointing, however, that the authors do not discuss another equally vital aspect: namely, the actual growth of networks, how patterns of symptoms evolve over developmental time.

In attempting to understand diseases, it is not suffcient merely to examine what their symptoms look like in an adult state. Particularly for deriving treatments and interventions, it is vitally important to track the progress of diseases throughout ontogeny to maturity. In doing so, a number of vital features can become apparent:

First, symptoms can fan out over time: that is, a small, basic-level deficit can, during development, lead to impairments in a variety of domains. So, for instance, an impairment in a hub cognitive domain such as executive attention (Cornish et al. Reference Cornish, Scerif and Karmiloff-Smith2007; Scerif et al. Reference Scerif, Cornish, Wilding, Driver and Karmiloff-Smith2005) can, if present during particular sensitive periods, impede development across a variety domains such as number, language, and other aspects of social development. Thus, one symptom can lead over developmental time to a range of other symptoms, a causal relationship that may be revealed only if the developmental perspective is considered.

Second, particular symptoms can fan in: that is, conditions can converge on a common pathway. One example here may be autism, where various authors (e.g., Anderson et al. Reference Anderson, Hooker and Herbert2008; Chauhan et al. Reference Chauhan, Chauhan and Brown2010; Herbert Reference Herbert2005) have hypothesised that a variety of different genetic vulnerabilities (i.e., discrete etiologies) may all converge on causing over-zealous neuroinflammatory responses early in neural development. These inflammatory and oxidative stressors early in development might, in turn, fan out (i.e., disrupt subsequent development in a range of ways), leading to many of the behavioural features that we recognise as the autistic spectrum behavioural phenotype. Again, it is only by adopting a developmental approach that such common pathways (or bottlenecks) may become evident; and yet they may offer potentially vital targets when developing treatment.

Similar principles of fan in and fan out can also go some way to explaining comorbidity, and to patterns of covariance of symptoms between different conditions. Thus, for example, any condition associated with impaired executive attention early in development (including, in various forms, Williams syndrome, Down syndrome, fragile X syndrome, and autism) may lead to partially similar patterns of impaired performance, which may then diverge again later in development (Cornish et al. Reference Cornish, Scerif and Karmiloff-Smith2007).

Another example of how developmental cascades operate can be gleaned from the neurodevelopmental disorder, Williams syndrome (WS; Donnai & Karmiloff-Smith Reference Donnai and Karmiloff-Smith2000). Infants and toddlers with WS are very impaired early on in planning saccadic eye movements (Brown et al. Reference Brown, Johnson, Paterson, Gilmore, Longhi and Karmiloff-Smith2003). This affects their subsequent ability to follow pointing (Laing et al. Reference Laing, Butterworth, Ansari, Gsodl, Longhi, Panagiotaki, Paterson and Karmiloff-Smith2002), which in turn is detrimental to their ability to use parental referential pointing to learn vocabulary. So, although their language becomes proficient (not “intact”; Karmiloff-Smith Reference Karmiloff-Smith1998 much later in development, initially language in toddlers with WS is extremely delayed and follows a deviant developmental trajectory (Annaz et al. Reference Annaz, Karmiloff-Smith, Thomas, Reed and Warner Rogers2008; Paterson et al. Reference Paterson, Brown, Gsodl, Johnson and Karmiloff-Smith1999). Thus, an early problem within the visual system, together with other contributing factors (Masataka Reference Masataka2001; Nazzi et al. Reference Nazzi, Paterson and Karmiloff-Smith2003), dynamically influences the way in which auditory stimuli are acquired and, because of the need for a critical vocabulary mass before syntax can take off, it is a visual deficit that is at the root of serious delays in grammatical development. Moreover, the failure to plan effcient saccadic eye movements doesn't only affect the learning of language. Individuals with WS also turn out to be predominantly featural processors, obvious from both brain and behavioural studies (Karmiloff-Smith et al. Reference Karmiloff-Smith, Thomas, Annaz, Humphreys, Ewing, Brace, Van Duuren, Pike, Grice and Campbell2004; Grice et al. Reference Grice, Spratling, Karmiloff-Smith, Halit, Csibra, de Haan and Johnson2001; Reference Grice, de Haan, Halit, Johnson, Csibra, Grant and Karmiloff-Smith2003; Mills et al. Reference Mills, Alvarez, St. George, Appelbaum, Bellugi and Neville2000). A possible explanation for this is that, in the typical case, rapid configural processing emerges from rapid scanning of stimuli via rapid eye movement planning whereas, in the atypical WS case, remaining fixated on a stimulus (e.g., a face) leads to a focus on featural detail. Individuals with autism are also featural analysers, yet the developmental pathway that leads to this end product may be different from the pathway that leads to featural processing in WS. Again, only a developmental approach, particularly using cross-syndrome comparisons, can reveal the dynamics of the differing routes from infancy to the mature state in adulthood. And associations across syndromes might be more informative than the search for dissociations (Karmiloff-Smith Reference Karmiloff-Smith2009; Karmiloff-Smith et al. Reference Karmiloff-Smith, Scerif and Ansari2003).

There is an increasing awareness that it is important to study not simply a “snapshot” of symptoms in the mature state, but also the process by which they were reached. In terms of network theory, this poses considerable but not (we think) insuperable demands on the modelling techniques that are used. Useful cross-discipline approaches may come from work on the development and evolution of social networking (e.g., Backstrom et al. Reference Backstrom, Huttenlocker, Kleinberg and Lan2006; Kumar et al. Reference Kumar, Novak and Tomkins2006), where plentiful data are available for all stages of network development, throughout ontogeny. Data from the very early phases of diseases (when the critical building blocks are laid, but before a clinical diagnosis can be made, as is often the case in autism, dyslexia, and dyscalculia) are often harder to come by, which is why conceptual insights from other areas of network science may be so useful.