Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Delayed stabilization and individualization in connectome development are related to psychiatric disorders

Abstract

The brain functional connectome constitutes a unique fingerprint allowing identification of individuals among a pool of people. Here we establish that the connectome develops into a more stable, individual wiring pattern during adolescence and demonstrate that a delay in this network tuning process is associated with reduced mental health in the formative years of late neurodevelopment.

This is a preview of subscription content, access via your institution

Relevant articles

Open Access articles citing this article.

Access options

Buy article

Get time limited or full article access on ReadCube.

$32.00

All prices are NET prices.

Figure 1: Preprocessing and subject identification procedure.
Figure 2: The brain functional connectome is more distinct with increasing age.
Figure 3: Individuals with increased clinical symptom scores showed a marked delay in connectome distinctiveness as compared to healthy individuals.

References

  1. Finn, E.S. et al. Nat. Neurosci. 18, 1664–1671 (2015).

    CAS  Article  Google Scholar 

  2. Paus, T., Keshavan, M. & Giedd, J.N. Nat. Rev. Neurosci. 9, 947–957 (2008).

    CAS  Article  Google Scholar 

  3. Satterthwaite, T.D. et al. Neuroimage 124 (Pt. B), 1115–1119 (2016).

    Article  Google Scholar 

  4. Power, J.D., Barnes, K.A., Snyder, A.Z., Schlaggar, B.L. & Petersen, S.E. Neuroimage 59, 2142–2154 (2012).

    Article  Google Scholar 

  5. Pruim, R.H. et al. Neuroimage 112, 267–277 (2015).

    Article  Google Scholar 

  6. Salimi-Khorshidi, G. et al. Neuroimage 90, 449–468 (2014).

    Article  Google Scholar 

  7. Shen, X., Tokoglu, F., Papademetris, X. & Constable, R.T. Neuroimage 82, 403–415 (2013).

    CAS  Article  Google Scholar 

  8. Davidson, R.J. Cerebrum 2014, 8 (2014).

    PubMed  PubMed Central  Google Scholar 

  9. Freund, J. et al. Science 340, 756–759 (2013).

    CAS  Article  Google Scholar 

  10. Caspi, A. & Moffitt, T.E. Nat. Rev. Neurosci. 7, 583–590 (2006).

    CAS  Article  Google Scholar 

  11. Fuhrmann, D., Knoll, L.J. & Blakemore, S.J. Trends Cogn. Sci. 19, 558–566 (2015).

    Article  Google Scholar 

  12. Insel, T.R. Nature 468, 187–193 (2010).

    CAS  Article  Google Scholar 

  13. Sawyer, S.M. et al. Lancet 379, 1630–1640 (2012).

    Article  Google Scholar 

  14. Patel, V., Kieling, C., Maulik, P.K. & Divan, G. Arch. Dis. Child. 98, 323–327 (2013).

    Article  Google Scholar 

  15. Ragland, J.D. et al. Neuropsychology 16, 370–379 (2002).

    Article  Google Scholar 

  16. Satterthwaite, T.D. et al. Neuroimage 61, 723–729 (2012).

    Article  Google Scholar 

  17. Gur, R.C. et al. Neuroimage 16, 651–662 (2002).

    Article  Google Scholar 

  18. Gur, R.E. et al. Arch. Gen. Psychiatry 64, 1356–1366 (2007).

    Article  Google Scholar 

  19. Biswal, B.B. Neuroimage 62, 938–944 (2012).

    Article  Google Scholar 

  20. Satterthwaite, T.D. et al. Neuroimage 86, 544–553 (2014).

    Article  Google Scholar 

  21. Smith, S.M. et al. Neuroimage 23 (Suppl. 1), S208–S219 (2004).

    Article  Google Scholar 

  22. Jenkinson, M., Bannister, P., Brady, M. & Smith, S. Neuroimage 17, 825–841 (2002).

    Article  Google Scholar 

  23. Pruim, R.H., Mennes, M., Buitelaar, J.K. & Beckmann, C.F. Neuroimage 112, 278–287 (2015).

    Article  Google Scholar 

  24. Roalf, D.R. et al. Neuroimage 125, 903–919 (2016).

    Article  Google Scholar 

  25. Fischl, B. et al. Neuron 33, 341–355 (2002).

    CAS  Article  Google Scholar 

  26. Jenkinson, M. & Smith, S. Med. Image Anal. 5, 143–156 (2001).

    CAS  Article  Google Scholar 

  27. Greve, D.N. & Fischl, B. Neuroimage 48, 63–72 (2009).

    Article  Google Scholar 

  28. Andersson, J.L.R., Jenkinson, M. & Smith, S. in FMRIB Technical Report TR07JA2 (FMRIB Centre, 2010).

  29. Fjell, A.M. et al. Neuroimage 50, 1376–1383 (2010).

    Article  Google Scholar 

  30. Biswal, B.B. et al. Proc. Natl. Acad. Sci. USA 107, 4734–4739 (2010).

    CAS  Article  Google Scholar 

  31. Poppe, A.B. et al. Cogn. Affect. Behav. Neurosci. 13, 641–659 (2013).

    Article  Google Scholar 

  32. Smith, S.M. et al. Proc. Natl. Acad. Sci. USA 106, 13040–13045 (2009).

    CAS  Article  Google Scholar 

  33. Beckmann, C.F. & Smith, S.M. IEEE Trans. Med. Imaging 23, 137–152 (2004).

    Article  Google Scholar 

  34. Filippini, N. et al. Proc. Natl. Acad. Sci. USA 106, 7209–7214 (2009).

    CAS  Article  Google Scholar 

  35. Ledoit, O. & Wolf, M. J. Empir. Finance 10, 603–621 (2003).

    Article  Google Scholar 

  36. Kaufmann, T. et al. Neuroimage 127, 324–332 (2016).

    Article  Google Scholar 

  37. Smith, S.M. et al. Neuroimage 54, 875–891 (2011).

    Article  Google Scholar 

Download references

Acknowledgements

The authors were funded by the Research Council of Norway (213837, 223273, 204966/F20, 229129, 249795/F20), the South-Eastern Norway Regional Health Authority (2013-123, 2014-097, 2015-073), the KG Jebsen Foundation, the European Commission 7th Framework Programme (602450, IMAGEMEND) and the NIH BD2K award (U54EB020403). Support for the collection of the data sets was provided by grant RC2MH089983 awarded to R. Gur and RC2MH089924 awarded to H. Hakonarson. All subjects were recruited through the Center for Applied Genomics at The Children's Hospital in Philadelphia.

Author information

Authors and Affiliations

Authors

Contributions

T.K. and L.T.W. conceived the study; T.K. analyzed the data with contributions from L.T.W.; D.A. N.T.D., C.L.B. and O.A.A. contributed with data preprocessing, quality assurance and interpretation of results; T.K. and L.T.W. wrote the first draft of the paper and all authors contributed to the final manuscript.

Corresponding authors

Correspondence to Tobias Kaufmann or Lars T Westlye.

Ethics declarations

Competing interests

The authors declare no competing financial interests.

Integrated supplementary information

Supplementary Figure 1 Residualized connectome distinctiveness (regressing motion, tSNR and motion × tSNR) depicted similar neurodevelopmental patterns.

Sex differences (a) and clinical patterns (b) in the relationship between age and motion as well as age and tSNR (upper plots). As expected and in line with the literature, motion estimated from raw data showed a strong decline with age in the current age-span (r=-0.35, p=5e-25). An ANCOVA on motion accounting for age, sex, gF (clinical symptom score) and their interactions revealed this strong effect of age (F=119, p→0) in addition to a significant effect of sex (F=8.8, p=.003) and a significant age*sex interaction (F=7.5, p=.007). There was no effect of gF (F=0.28, p=.6) nor a significant interaction involving gF (all p>.47). A similar ANCOVA on the resulting tSNR after cleaning showed a significant age*sex interaction (F=6.7, p=.01) and a sex effect close to nominal significance (F=3.5, p=.06), yet all other effects were clearly non-significant. Thus, these motion and tSNR patterns cannot easily explain the patterns observed in connectome distinctiveness (lower plots, residuals after regressing motion, tSNR and motion*tSNR). The curves in the lower plots represent mean (solid line) and standard deviation (shaded areas) of smooth function fitting across 10,000 bootstraps.

Supplementary Figure 2 Excluding subjects such that the resulting motion profiles were similar between sexes yielded similar sex patterns in connectome distinctiveness.

Ruling out that differences in the motion profiles of boys and girls could explain the patterns in connectome distinctiveness we have removed individuals based on mean motion as well as based on the standard deviation in motion across runs (a), such that the retaining motion patterns were similar across sexes (b), i.e. the remaining individuals moved little and displayed similar motion patterns across runs. The resulting pattern of sex differences in connectome distinctiveness from this subgroup (c) matches the one reported based on the full sample and we conclude that motion differences cannot easily explain the reported neurodevelopmental differences between males and females. The curves in (c) represent mean (solid line) and standard deviation (shaded areas) of smooth function fitting across 10,000 bootstraps.

Supplementary Figure 3 Excluding subjects such that the resulting motion profiles were similar between HC and gF yielded similar clinical patterns in connectome distinctiveness.

Ruling out that differences in the motion profiles between healthy controls and individuals with increased gF scores could explain the clinical patterns in connectome distinctiveness, we have removed individuals based on mean motion as well as based on the standard deviation in motion across runs (a), such that the retaining motion patterns were similar across groups (b), i.e. the remaining individuals moved little and displayed similar motion patterns across runs. The resulting pattern of clinical differences in connectome distinctiveness from this subgroup (c) matches the one reported based on the full sample and we conclude that motion differences cannot easily explain clinical differences in connectome distinctiveness. The curves in (c) represent mean (solid line) and standard deviation (shaded areas) of smooth function fitting across 10,000 bootstraps.

Supplementary Figure 4 Subject identification based on motion and tSNR estimates alone (no connectome data) did not succeed.

We ruled out that the basic findings can be explained by a model based on motion alone, by performing the subject identification procedure purely based on motion and tSNR estimates (no connectome data). For each individual and run we extracted the average across volumes for the 6 motion parameters from FSL MCFLIRT (3 rotation, 3 transformation). In addition, we extracted 1 estimate of mean relative frame-to-frame displacement from the raw data and an estimate of tSNR from the AROMA+FIX cleaned data. This resulted in 8 estimates of subject-specific motion/tSNR parameters for each run (WM, ER, RS) based on which we attempted to identify individuals using similar procedures as we used for the connectome in the main text. The resulting identification accuracies were extremely low (a), in fact, 96.1% of subjects were not identified correctly in any of the identification tasks. The resulting QC-based distinctiveness did not display a significant association with age or sex (b) and the resulting neurodevelopmental trajectories (c-d) did not match the observed patterns in connectome distinctiveness. P-values in (b) were obtained through permutation testing across 10,000 permutations. The curves in (c-d) represent mean (solid line) and standard deviation (shaded areas) of smooth function fitting across 10,000 bootstraps.

Supplementary Figure 5 The reported age trajectories in connectome distinctiveness were reproduced using a different analysis pipeline based on independent component analysis (ICA).

To verify that our results did not depend on the choice of the functional whole brain atlas, we have performed similar analyses decomposing the fMRI signal of all runs (three per participant: WM, ER and RS) into spatially independent components and computing functional connectivity matrices by means of regularized partial correlations of the time series. The corresponding results match well those reported in the main paper. Of note, these results also speak against a dependence of main results on motion confounds, both since the specific approach involves an additional step of noise regression, and because ICA will often model characteristic noise signals in their own components. (a) Resulting accuracy (b) T-statistics from a linear model. P-values were obtained through permutation testing across 10,000 permutations. (c-d) Estimated trajectories. The curves represent mean (solid line) and standard deviation (shaded areas) of smooth function fitting across 10,000 bootstraps.

Supplementary Figure 6 Comparison of statistical models calculated in extreme groups (group differences) or across all subjects (continuous associations with symptom scores).

Models corresponding to the group difference patterns as visualized in Fig. 3 (a) and models corresponding to continuous associations with symptom scores (b; no group selection bias). P-values were obtained through permutation testing across 10,000 permutations. Significant associations (accounting for the number of networks) are indicated with white font and white boxes.

Supplementary Figure 7 With increasing age, the intrinsic network architecture is stabilized across runs.

The brain functional connectome grounds on an intrinsic network architecture that is common to resting state and task MRI and shares profound similarities across individuals (Cole et al., 2014; Kaufmann et al., 2017). (a) Average connectivity matrix of RS, WM and ER in the PNC sample. Clearly, the patterns overlap across contexts (r>.94 for all pairwise correlations of average connectivity profiles). (b) Comparison to average resting state connectivity from 276 healthy adults from one of our own in-house samples. The overlap between the connectome from PNC and the comparison sample is visually evident (r>.82 for all pairwise correlations between PNC and comparison sample) and support the existence of a common, intrinsic network architecture, general across subjects and contexts. (c) Maturation patterns in connectome development, splitting the PNC sample into six age cohorts, 133 subjects each, and averaging resting state connectivity across subjects of each cohort. (d) Maturation patterns in the variance across runs (RS, WM and ER) for each age cohort. Together, (c-d) suggest a maturation pattern in the intrinsic network architecture toward a connectome backbone that is more prominent with increasing age and more coherent across tasks and contexts.

Cole, M.W., Bassett, D.S., Power, J.D., Braver, T.S. & Petersen, S.E. Neuron 83, 238-251 (2014).

Kaufmann, T., et al. Neuroimage 147, 243-252 (2017).

Supplementary Figure 8 Time series split analysis confirms that the increased across-context coherence with increasing age is unlikely to be a side effect of differential noise profiles between young and old subjects.

(a) We performed an analysis, where we split the time series of each fMRI run into two segments and computed connectivity for each of the two segments per run and subject. If reliability of networks was the key explanation of an increased across context stability, we would expect a strong age pattern in the within-subject correlations of connectivity of the splits. However, the results in (b) show high correlations between connectivity profiles of both splits and indicate no strong age associations (all r<=0.08; for comparison, the association between across-context stability and age is r>.43, p<2e-37 in all combinations of runs). Thus, these results support that the increased across-context coherence with increasing age is unlikely an artifact of measurement but rather suggest that it resembles biologically interpretable, neurodevelopmental changes.

Supplementary Figure 9 Testing two hypotheses, one favoring an individual-level tuning and one favoring a common tuning, the results provide strong support of a network tuning with less overlapping (more individualized) edges with increasing age.

We tested on the level of network edges if an increased across-context stability with increasing age yields individuality. The conceptual drawing in (a) illustrates the hypothesized development of two subjects (red and blue) observed at two time points in neurodevelopment (child- and adulthood), where the error bars would depict the across-context variation in connectivity strength of a given edge. Both plots show a decrease in across-context variation with increasing age (smaller error bars in older subjects), and the mean connectivity value is more similar across subjects with increasing age, i.e. both patterns match the true patterns observed in Supplementary Fig. 7. However, whereas the upper plot yields non-overlapping edges in old individuals (individual level tuning resulting in edges that are stabilized around an »individual« edge strength), the lower plot does not (increasing across-context coherence without an individual level tuning, stabilizing edges around a »common« edge strength). We tested which of these two hypotheses is more likely, computing how much the error bar (variation across runs) of each subject overlaps with the error bar of all other subjects (percent of overlap). Almost all edges in the brain showed a clear reduction in the average percentage of “overlap with others” with increasing age (b), clearly supporting that in the process of stabilizing the connectome across runs, an individual level tuning of connectivity strength manifests, yielding a connectome that is robustly different from others, independent of task-induced modulations of connectivity strength. These results therefore provide evidence on the edge level for an increased individuality with increasing age.

Supplementary Figure 10 Multivariate estimates of stability and individuality were strongly related.

After assessing the edge level in Supplementary Fig. 9, we also tested on the network level if an increased across-context stability with increasing age yields individuality. We utilized distance metrics (one minus the Spearman’s rank correlation coefficient) and inferred individuality as the minimum distance between any of one’s own connectomes and any other connectome in the sample divided by the maximum distance between one’s own connectomes (i.e. between RS, WM and ER). We refer to this as the distance ratio:

Next, we calculated each subject’s average correlation of connectivity profiles across runs as a pure measure of across-context stability and compared it to dratio. Individuality (indexed by dratio) and stability were strongly linked in a non-linear relationship, suggesting that dratio is particularly boosted after a certain level of across-context stability is reached. This complies well with the results from the edge level, in that also on the network level an increasing across-context stability may yield individuality.

Supplementary Figure 11 Results from statistical models on connectome distinctiveness and gF in the various networks confirm a contribution of both stability and individuality.

(a) Results from linear models associating connectome distinctiveness with dratio and across-context stability (see Supplementary Fig. 10 for an explanation of the measures), covarying for age, sex, gF, mean motion, sd motion and tSNR (left plot) and from linear models associating gF with dratio and across-context stability, covarying for age, sex, mean motion, sd motion and tSNR (right plot). The resulting t-scores for dratio and stability are depicted. (b) One could argue that dratio might be confounded by across-context stability in that distances obtained within subjects across runs are part of the equation. Therefore, we have also run similar analysis where we only entered the numerator of dratio as an index of individuality, i.e. the minimum distance to all other subjects’ connectomes. Both panels suggest significant contributions of both individuality and stability to connectome distinctiveness and moderate associations with gF. P-values were obtained through permutation testing across 10,000 permutations. Significant associations (accounting for the number of networks) are indicated with white font in white boxes, associations significant at a nominal level (p<.05) are indicated with red boxes.

Supplementary Figure 12 Age distribution for each clinical group in comparison to healthy controls.

Supplementary Tables 1 and 2 provide additional detail on group formation and sample sizes.

Supplementary information

Supplementary Text and Figures

Supplementary Figures 1–12 and Supplementary Tables 1 and 2 (PDF 2535 kb)

Supplementary Methods Checklist (PDF 566 kb)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Kaufmann, T., Alnæs, D., Doan, N. et al. Delayed stabilization and individualization in connectome development are related to psychiatric disorders. Nat Neurosci 20, 513–515 (2017). https://doi.org/10.1038/nn.4511

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/nn.4511

Further reading

Search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing