In their timely article, Genon and colleagues review recent developments in MRI studies that aim to link human behaviour to brain structure (Genon, S., Eickhoff, S. B. & Kharabian, S. Linking interindividual variability in brain structure to behaviour. Nat. Rev. Neurosci. 23, 307–318 (2022))1. Over the past decade, they argue, the field has witnessed low replicability of research findings and decreases in effect sizes. They point to the adoption of multivariate approaches as one promising path forward. We endorse their insightful suggestions and would like to draw attention to two additional points that, in our opinion, represent key challenges and possible solutions going forward.
There exists no simple one-to-one relationship between a structural MRI signal and the underlying ‘true’ brain structure. MRI signals reflect mixed contributions from various structural components within a voxel, and some of these components affect brain functions in markedly different ways. For example, an increase in a quantitative T1 MRI signal can result from decreased myelination or increased axon diameter2,3 (Fig. 1a), which affect signal conduction speed in opposite directions4. The gap between the MRI signal and the underlying brain structure poses a considerable challenge to brain structure–behaviour mapping.
Notably, some promising developments are underway to bridge this gap. We want to highlight two such developments: multidimensional and multimodal MRI5. By acquiring multiple structural MRI signals, with each signal reflecting a different weighted sum of structural components, these techniques can disentangle and measure individual structural components such as myelin level6, axon diameter7 and cell morphology8. The measures represent functionally more relevant units of the brain and provide opportunities for mechanistic insights.
Another challenge to brain structure–behaviour mapping is the many-to-one relationship between brain structure and behaviour. As Genon and colleagues noted1, the field has long relied on the assumption of a linear structure–behaviour relationship. However, recent studies have raised doubts about this assumption, pointing instead towards a many-to-one structure–behaviour relationship, known as ‘multiple realizability’. For example, a U-shaped relationship was observed between visual performance and visual cortical volume, suggesting that the degradation of visual performance can result from increased cortical thickness or decreased cortical surface area9 (Fig. 1b). Likewise, a many-to-one relationship exists between network structure and network behaviour10.
The lack of a one-to-one relationship between brain structure and behaviour adds an important reason for adopting multivariate and machine learning approaches. These approaches can inspect the entire space of structure–behaviour relationships with minimal pre-assumption. A promising application of these approaches, we suggest, is the search for optimal brain structure. It provides opportunities to address what ratio of myelin to axon is optimal for signal conduction, what ratio of white to grey matter is optimal for different domains of behaviour, and other conceptually important questions.
In summary, the field is challenged, in our opinion, by a lack of one-to-one mapping from MRI to brain structure and from brain structure to behaviour (Fig. 1). Progress, therefore, relies largely on the ability to bridge the gap from MRI to brain structure and to examine the multiple realizability of behaviour in brain structure. Recent developments along these lines, such as advanced MRI techniques and advanced statistical approaches, provide opportunities for a better conceptual understanding of how multifaceted human behaviour emerges from human brain structure.
Genon, S., Eickhoff, S. B. & Kharabian, S. Linking interindividual variability in brain structure to behaviour. Nat. Rev. Neurosci. 23, 307–318 (2022).
Harkins, K. D. et al. The microstructural correlates of T1 in white matter. Magn. Reson. Med. 75, 1341–1345 (2016).
Lutti, A., Dick, F., Sereno, M. I. & Weiskopf, N. Using high-resolution quantitative mapping of R1 as an index of cortical myelination. Neuroimage 93, 176–188 (2014).
Drakesmith, M. et al. Estimating axon conduction velocity in vivo from microstructural MRI. Neuroimage 203, 116186 (2019).
Tax, C. M. W. in New Developments in NMR (ed. Topgaard, D.) Ch. 7, 186–227 (Royal Society of Chemistry, 2020).
Deoni, S. C. L., Rutt, B. K., Arun, T., Pierpaoli, C. & Jones, D. K. Gleaning multicomponent T1 and T2 information from steady-state imaging data. Magn. Reson. Med. 60, 1372–1387 (2008).
Assaf, Y., Blumenfeld-Katzir, T., Yovel, Y. & Basser, P. J. AxCaliber: a method for measuring axon diameter distribution from diffusion MRI. Magn. Reson. Med. 59, 1347–1354 (2008).
Palombo, M. et al. New paradigm to assess brain cell morphology by diffusion-weighted MR spectroscopy in vivo. Proc. Natl Acad. Sci. USA 113, 6671–6676 (2016).
Song, C., Schwarzkopf, D. S., Kanai, R. & Rees, G. Neural population tuning links visual cortical anatomy to human visual perception. Neuron 85, 641–656 (2015).
Marder, E., Goeritz, M. L. & Otopalik, A. G. Robust circuit rhythms in small circuits arise from variable circuit components and mechanisms. Curr. Opin. Neurobiol. 31, 156–163 (2015).
The authors received funding from the Wellcome Trust (209192/Z/17/Z to C.S.), H2020 MSCA COFUND (663830-CU119 to C.S.), the Danish Foundation for Research in Neurology (to K.S.) and OPUS (grant 2017/27/B/HS6/00937 to K.S. and R.R.). In addition, the contributions of K.S. and R.R. to this article were based upon work from COST Action CA18106, supported by COST (European Cooperation in Science and Technology).
The authors declare no competing interests.
About this article
Cite this article
Song, C., Sandberg, K., Rutiku, R. et al. Linking human behaviour to brain structure: further challenges and possible solutions. Nat Rev Neurosci 23, 517–518 (2022). https://doi.org/10.1038/s41583-022-00614-4