Long-term neural and physiological phenotyping of a single human (MyConnectome)

Nature Communications 6:8885, doi 10.1038/ncomms9885; open access (CC BY 4.0), PMC4682164. Read for research direction R1 (long-duration and dense recording of one person), 4 October 2026. Provenance: papers/amadeus/poldrack2015_myconnectome.provenance.json.

What was read

• Read in full: the Europe PMC full-text XML converted to text (abstract, main text, Tables 1–2, all figure captions, all Methods, accession codes, acknowledgements, contributions, references), and the 17-page supplement from the Europe PMC supplementary bundle (Supplementary Tables 1–2, gene and metabolite modules; Supplementary Note 1, every survey question).
• Not read: the figures as images (captions only). Laumann et al. 2015 (Neuron), which holds the detailed resting-state analyses this paper builds on, was not read.

Design

• One person: the first author, a healthy right-handed man aged 45 at the start, with plaque psoriasis and a history of anxiety disorder. Self-experimentation. The University of Texas ruled that it was not human-subjects research, so there was no IRB there; the Washington University session ran under an IRB protocol.
• Duration: pilot from 25 September 2012; production phase 22 October 2012 to 11 March 2014 (532 days, with a hiatus from 6 March to 30 April 2013); follow-ups in April and May 2015. 104 scanning sessions, stopping rule fixed by sessions and blood draws (target 100 scans, then 48 blood draws), set before any time-series analysis.
• Schedule: fixed times, Mondays 17:00 and Tuesdays and Thursdays 07:30 (Monday sessions dropped after the hiatus). Tuesdays were fasted and caffeine-free because of the blood draw that followed the scan.
• MRI (3T Skyra, 32-channel coil):
  • Resting state: 100 runs of 10 minutes, 84 usable (14 h acquired, about 12.6 h after dropping the first minute of each run): multiband EPI, TR 1.16 s, 2.4 × 2.4 × 2 mm. Active noise cancellation, which sometimes failed.
  • Tasks: n-back (15), dot-motion stop signal (8), object localiser (8), spatial working memory (4), language localiser (5), retinotopy (1), breath-holding (18).
  • Structure: T1 (21 acquisitions, 10 usable), T2, diffusion (19 sessions, 15 usable), one HARDI session.
  • Follow-up: ten eyes-open and ten eyes-closed 10-minute runs at another site.
• Biology: weekly blood after the Tuesday scan, giving 48 RNA-seq samples (13,072 genes after filtering) and 48 metabolomics samples (106 named metabolites); 23andMe genotyping; whole-exome sequencing.
• Self-report and behaviour (the labels):
  • Morning: sleep quality, soreness, blood pressure, pulse.
  • After each scan: the PANAS-X affect list; anxiety during the scan; and a free-text "Thoughts during scan".
  • Evening: time outdoors, psoriasis, stress, gut health, a free-text end-of-day report, exercise, alcohol, and all food eaten.
  • Other streams: weight from a Fitbit Aria scale; sleep from a ZEO headband on most nights before scans ("data were missing for a number of nights due to the sensor falling off or sensor failure"); daily weather; sent e-mails scored with LIWC.
• Analysis:
  • Brain: an individual parcellation (616 cortical plus 14 subcortical regions) built from all sessions concatenated, and 12 Infomap networks.
  • Statistics: time series related to each other bivariately with ARIMA models chosen automatically (autocorrelation was substantial), FDR q < 0.1 with |r| > 0.2. Phenome-wide: 38,363 tests, Bonferroni.
  • Bias control: the subject was blind to analyses for about the first six months; a priori hypotheses were recorded after the hiatus.

Main results (verified)

• Networks were largely stable over 18 months. Visual, somatomotor and dorsal attention networks varied most from session to session. The authors note this is the reverse of between-person studies, where primary sensory and motor regions vary least across people.

  • Some networks showed linear or polynomial trends; the authors say these could reflect habituation or sensitisation to repeated scanning.
  • Whole-connectome similarity to the mean had no long-range trend, but sessions formed clusters in time. Modularity and global efficiency were stable.

• Physiological state changed the networks. Fasted, caffeine-free Tuesdays against fed Thursdays: greater within-network connectivity in somatomotor, dorsal attention and primary visual networks, and a reorganised hub structure. The authors warn that fasting and caffeine are confounded.

• Task and rest connectomes correlated at r = 0.45 (237 task contrasts).

• Exploratory brain–behaviour–biology links. The strongest Bonferroni-surviving links were:

  • post-scan fatigue with visual–cingulo-opercular connectivity, r = 0.621 (N = 74);
  • fed/fasted day with somatomotor–medial-parietal connectivity, r = 0.532;
  • several metabolites and gene-expression modules with between-network connectivity (r about 0.35–0.57, N = 39).

  A phenome-wide network had modularity 0.70. One module joined mood, connectivity within several networks, fatigue and fed/fasted state; another joined lipid metabolism, three networks and sleep quality.

• Mood and immune gene expression were unrelated, against a recorded a priori hypothesis. Psoriasis severity was related to T-cell-receptor expression.

• Power for follow-up studies: 24 to 51 sessions for 80% power to replicate connectivity–gene-expression associations of |r| 0.54 to 0.38.

• Burden and safety:

  • Tinnitus rose during the pilot. Audiometry showed high-frequency loss, likely pre-existing. A worsening on 6 March 2013 (to 70 dB HL at 6 kHz) stopped imaging for about two months. The protocol restarted limited to noise-cancelled acquisitions; the loss was not sustained.
  • The authors judge that healthy volunteers are "unlikely to be sufficiently motivated to sustain frequent participation" and that self-experimentation "was the only means by which this proof-of-concept study could have been completed."

• Data: raw MRI on OpenfMRI ds031 (now OpenNeuro ds000031), RNA-seq on GEO GSE58122 and SRA SRP042596, code and a virtual machine on GitHub. "The complete imaging, genomic and behavioural data are made openly available with no restrictions on access/usage."

Limits

• One person, who was also the principal investigator. Expectation effects cannot be excluded, and the authors say so; blinding was partial and ended after the hiatus.
• Exploratory discovery analyses, bivariate with uneven missingness. Associations are hypotheses, not findings, and most have N of 39 to 84 time points.
• Rest consisted of 10-minute runs two to three times a week at fixed hours. Nothing was recorded continuously, and nothing outside the scanner was neural (sleep came from a consumer headband with gaps).
• The free-text "thoughts during scan" was collected but not analysed in the paper. No content of thought is decoded or related to the brain data.

What it means for Amadeus

• Feasibility (verified): one healthy, motivated adult can be scanned about 100 times over a year and a half with daily questionnaires, weekly blood and consumer sleep sensing. The project's own authors doubt that ordinary volunteers would sustain it. Scanner noise is a real cumulative risk: audiometry has to be part of any dense-scanning design.
• What dense rest scanning returns (verified): mainly a stable, person-specific network map, plus measurable state effects (fasting and caffeine, fatigue). The brain–behaviour relations are with bodily and affective state, not with what the person thinks or remembers. A month of repeated rest scans would mostly re-measure the same fingerprint more precisely. It would not record thought content (inference, consistent with Gratton 2018 in this folder).
• Labels (verified): experience sampling here was coarse, with once or twice daily questionnaires, plus a free-text report of thoughts after each rest scan. That post-scan free text is the cheapest form of the "what were you just thinking?" label of brief 6.7, but it is retrospective over ten minutes and was never analysed.
• E3 (verified for terms; content of the release not checked): fully open data on one person over 18 months, pairing rest fMRI with same-day and next-day self-report and behaviour. It is not autobiographical recall. Whether the released behavioural tables include the free-text thought reports needs a metadata check before any analysis.
• Ethics (verified): this landmark single-person study ran without IRB review at its main site because the subject was the investigator. That route is not available for Amadeus with a participant other than the experimenter.