Over the past fifteen years, much has been written about various hazards of clinical neuroscience’s core methodologies, including low reproducibility and statistical power, artifact, confounding, and so on. Odysseus-like, we may press on with our challenging scientific journey through these hazards, aided by this growing literature. Nevertheless, there will always remain the question of the journey’s conclusion in Ithaca, and whether any new discoveries can be recognized, understood, and even welcomed. While theory is the basic currency of scientific comprehension, there is growing acknowledgment that some areas of neuroscientific investigation have been systematically under-evaluated experimentally. Sex differences and sex-specific neural and psychological processes are one such example [1], and a paucity of evidence hinders theory construction in this area.

Furthermore, studying the neuroscience of human childbirth and motherhood is particularly difficult. Evidence points towards a reliable elevation of the risk of major depression onset at least 0–3 months post-partum, and potentially longer [2], as well as possible interactions with other psychiatric disorders [3]. This window of elevated risk is suggestive that childbirth, and its sequelae, have a causal effect on psychiatric outcomes. However, the tangle of potential candidate mechanisms [3], including changes to hormone levels, neural structure and function, stress, inflammation, and to the mother’s everyday life and behavior, do not yield straightforwardly to a simple account, nor can easily be evaluated using many forms of typical experimental design. Thus, a reductive approach to understanding the contribution of one brain region or neurotransmitter to post-partum behavioral changes is unlikely to be feasible, could be limited by ethical concerns, and/or may miss the point.

The study of Lotter and colleagues [4], while providing new information into neural and affective changes following childbirth, also provides broader insights into how to approach highly complex neuroscientific questions of this sort. The authors collected data in 75 participants within 7 days post-partum (PP group), from which a subgroup of 19 participated in a longitudinal protocol with data collection at 3-, 6-, 9-, 12-, and 24-week timepoints post-partum. The baseline data were compared with a sample of 23 nulliparous women (NP). Measures included resting-state functional magnetic resonance imaging (rsfMRI) data, hormone levels (estradiol/progesterone), and self-report measures of mood. The rsfMRI data were analyzed using three different methods: first, fractional amplitude of low-frequency fluctuations (fALFF), which reflects the relative contribution of low-frequency neural activity, and may relate to integration of information across time; global connectivity (GCOR), which reflects a given voxel’s relationship with global neural signals, and hence its role in long-range/trans-network functional connectivity; local connectivity (LCOR), which reflects a voxel’s integration with other nearby (i.e. 6mm full-width, half-maximum kernel) voxels. The design allowed the identification of candidate neural differences between PP and NP groups at baseline, whose trajectories were then examined within the longitudinal sample. The potential for selection bias and regression-to-the-mean effects was evaluated.

I will pick several findings to highlight, somewhat arbitrarily: first, reduced fALFF in PP relative to NP within the right prefrontal cortex which was relatively stable across time; second, reduced GCOR in the bilateral putamen in PP relative to NP, which diminished across time; third, reduced LCOR in the bilateral insula in PP relative to NP, which was relatively stable across time; fourth, a spatial colocalization of PP versus NP (PP > NP) whole-brain fALFF differences with the Allen Brain atlas’s map of progesterone receptor density, which was also relatively stable across time; fifth, spatial colocalization of NP versus PP (NP > PP) whole-brain GCOR differences with progesterone receptor density, which appeared to diminish numerically across time but not significantly so; sixth, associations across time between putamen GCOR with blood progesterone concentration; seventh, associations across time of depression symptoms with the insula’s LCOR.

While it is tempting to interpret each finding in turn, a broader perspective may also be insightful: the number and variety of observations are consistent with the notion that childbirth is accompanied by complex patterns of neurobiological changes with different temporal dynamics [3]. Of these, some appear related to hormone concentrations, receptor density, or mood, but formal independence would require confirmation with larger samples. Although it remains possible that a simple causal mechanism may be identified among this panoply of changes, the scope of the findings suggests an alternative approach may be productive. Taking a dynamical systems perspective, van der Leemput and colleagues [5] observed multidimensional self-report measures of affect appear to increase in variability and become more strongly (cross-/auto-) correlated in the time leading up to the onset of depression. This pattern suggests that emotional reactivity may become less rapidly responsive to the dynamics and details of day-to-day external events, and rather starts to exhibit a longer-term momentum driving the individual towards a change in affective state.

One can imagine this type of model applying to post-partum psychiatric risk, with changing neural and peripheral physiological processes, interacting with stress, to interfere with emotion regulation, and potentially also maternal behavior. Testing this type of model would be challenging, but the study of Lotter and colleagues [4] shows how it might be achieved: detailed collection of subjective mood, and hormone levels as well as neurobiological markers across time, with information from multiple measures being needed to achieve prediction of psychiatric outcomes, perhaps in a non-linear fashion. Moreover, the use of GCOR, and the evaluation of distributed patterns of neural activity through relationships with receptor expression across the whole-brain, is, loosely speaking, analogous with the dynamical systems perspective. How far this analogy can be pursued remains to be seen but, at least, a global characterization of brain function may complement the regionally-localized approach by characterizing how individual nodes relate to global neural activity. More specifically, it may be insightful for understanding the higher-order statistical properties of affective dynamics [5], which presumably reflect the regulation of affect by distributed neural systems.

Even in the most empirically-oriented work, theory’s importance remains implicit throughout. It motivates and frames the endeavor, and gives new observations their significance by defining the scope and limitations of received wisdom. Thus, empiricism represents a transition between a previous conception of the world and a new one to be born.