Pith. sign in

REVIEW 3 major objections 5 minor 12 references

Functional Architecture of the Human Hypothalamus: Cortical Coupling and Subregional Organization Using 7-Tesla fMRI

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Seven-tesla resting-state fMRI reveals four functional subregions in the human hypothalamus, each with its own pattern of coupling to the cerebral cortex.

desk verdict A useful first 7T hypothalamic parcellation, but the network-level ANOVA is circular and the parcellation lacks validation before it can be treated as a reference. read the letter →

arxiv 2506.06191 v2 pith:CMCU2CET submitted 2025-06-06 q-bio.NC q-bio.QM

Kent M. Lee , Joshua Rodriguez , Ludger Hartley , Philip A. Kragel , Lorena Chanes , Tor D. Wager , Karen S. Quigley , Lawrence L. Wald
show 4 more authors
Marta Bianciardi Lisa Feldman Barrett Jordan E. Theriault Ajay B. Satpute
This is my paper
classification q-bio.NCq-bio.QM
keywords humanhypothalamus7TeslafMRIfunctionalparcellationresting-stateconnectivitycorticalcouplingdefaultmodenetworkallostaticcommunitydetection
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper aims to establish that the human hypothalamus, despite its small size and poorly defined internal boundaries, is not functionally one piece: resting-state scans at 7 tesla resolve it into four subregions defined by their intrinsic connectivity. This matters because the hypothalamus regulates metabolism, body temperature, feeding, and survival behaviors, while almost all prior human fMRI work treated it as a single region or imposed coarse lateral and medial divisions. The authors report that all four subregions couple more strongly with a frontal, midline, and limbic cortical network than with a posterior sensorimotor network, with the anterior subregion showing the strongest cortical coupling and an anteroventral-tuberal subregion showing the weakest. If right, this gives future studies concrete seeds for asking how distinct hypothalamic zones support different bodily-regulation functions in humans.

What carries the argument

The machinery is intrinsic functional connectivity parcellation at ultra-high field. For each participant, voxel-level BOLD time series inside a probabilistic mask of the hypothalamus are cleaned by general-linear-model nuisance regression and physiological noise correction, then correlated across the 1,019 hypothalamic voxels. A group-level connectivity matrix is submitted to Louvain community detection, an algorithm that groups voxels into modules whose members are more correlated with one another than with outsiders, and the clustering is repeated 100 times and re-clustered on a co-assignment matrix to stabilize the solution. The resulting four communities become seed regions whose averaged time series are correlated with 360 cortical parcels from a standard cortical atlas, and k-means clustering of the resulting connectivity profiles, with cluster number chosen by a cluster-validity index and an elbow rule, collapses the cortex into two networks. This two-stage procedure is what moves from clusters of hypothalamic voxels to subregion-specific cortical coupling patterns.

What would settle it

Re-analyze the same data under a null model that randomly breaks the spatial relationships among hypothalamic voxels while preserving each voxel's time course; if coherent four-community solutions appear as often in the null data as in the real data, the parcellation is indistinguishable from noise.

Watch

Extended reading notes

Core claim

The central discovery is a data-driven functional parcellation of the human hypothalamus. Analyzing BOLD time-series correlations among 1,019 hypothalamic voxels at 1.1 mm isotropic resolution in 104 participants, with a community-detection algorithm applied to the group-level connectivity matrix and repeated 100 times for stability, yields four spatially coherent communities: anteroventral-tuberal, anterior, middle tuberal-posterior, and superior. These communities do not show the lateral-to-medial split that older models assumed; they instead follow an anterior-posterior and dorsal-ventral arrangement. When each subregion's time series is correlated with 360 cortical parcels and those parcels are clustered by their connectivity profiles, two cortical networks emerge: one composed mainly of frontal, midline, and limbic areas, and another composed largely of posterior sensorimotor and dorsolateral prefrontal areas. All four hypothalamic subregions connect more strongly to the first cortical network, the anterior community strongest of all, while the anteroventral-tuberal community shows the weakest cortical connectivity overall and no significant coupling to the second network.

Load-bearing premise

The result depends on the assumption that the leftover signal in the hypothalamic voxels after noise removal reflects true neural coupling, rather than pulse, breathing, or fluid from the nearby third ventricle.

Editorial extensions

If this is right

  • Future human fMRI studies should not treat the hypothalamus as a single seed, because whole-structure or coarse medial-lateral seeds can average away subregion-specific coupling.
  • The four communities provide ready-made regions of interest for task studies of appetite, thermoregulation, sleep, and stress, allowing tests of whether anterior versus tuberal zones dissociate behaviorally.
  • The anterior community's strong coupling with prefrontal, cingulate, and insular cortex aligns with animal work showing that these cortical areas project to preoptic and anterior hypothalamic nuclei.
  • The weak cortical coupling of the anteroventral-tuberal community implies that its metabolic and neuroendocrine functions may be organized mainly through subcortical and humoral pathways rather than direct cortical loops.
  • The absence of a lateral-medial split suggests that human hypothalamic functional organization is better described by anterior-posterior and dorsal-ventral axes, in line with modern structural MRI parcellations.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural next test, not run here, is whether the four communities replicate in a split-half sample or under a different clustering algorithm; if the specific boundaries shift, the stable element may be the overall anterior-posterior gradient rather than the exact four-zone solution.
  • Because resting-state coupling is correlational, the communities may reflect shared physiological inputs rather than discrete anatomical nuclei; combining 7T fMRI with post-mortem tract tracing or high-resolution structural imaging would test that reading.
  • The paper's thermoregulation speculation could be tested directly by scanning participants during mild warming and cooling and asking whether anterior hypothalamic coupling to prefrontal, cingulate, and insular cortex changes more than tuberal coupling.
  • A control analysis regressing cerebrospinal-fluid signal or estimating partial-volume contamination would clarify whether the anterior community's broad, strong coupling is partly driven by proximity to the third ventricle.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The manuscript reports a data-driven, voxelwise parcellation of the human hypothalamus from 7T resting-state fMRI in 104 participants. Louvain community detection on a group-averaged intrinsic connectivity matrix within a 1019-voxel hypothalamic mask yields four communities (anteroventral-tuberal, anterior, middle tuberal-posterior, superior). The authors then parcellate 360 cortical Glasser parcels into two networks via k-means on their Fisher-z connectivity profiles to the four hypothalamic communities, and test network-by-community differences with a 2x4 repeated-measures ANOVA. They report that all hypothalamic communities show stronger connectivity with an anterior frontal/midline/limbic network than with a posterior sensorimotor network, with the anterior hypothalamic community strongest and the anteroventral-tuberal community weakest. The paper interprets these results as the first data-driven functional connectomic map of human hypothalamic subregions.

Significance. The target result, if secured, would be a valuable contribution: a high-resolution, in vivo, data-driven description of hypothalamic subregional organization and its cortical coupling, with potential utility as a resource for future seed-based studies. The study has notable strengths: a relatively large 7T sample, 1.1-mm isotropic acquisition, a consensus-based Louvain procedure, use of the Glasser cortical atlas, and a direct visual comparison with an anatomical parcellation. The main inferential claims, however, rest on two validation steps that are currently missing: an independent confirmation of the Louvain partition against noise or split-half data, and a cortical-network definition that is not circular with the ANOVA used to test it. Because both gaps are addressable with additional analyses, I view the manuscript as promising but not yet established.

major comments (3)
  1. [Results, 'Cortical Connectivity with Hypothalamic Subclusters'; Methods, 'Functional Connectivity Analysis'] The two cortical networks are defined by k-means on the 360 x 4 matrix of Fisher-z connectivity values between cortical parcels and hypothalamic communities, and the same values are then averaged within clusters and submitted to a 2x4 ANOVA. Because k-means explicitly maximizes between-cluster separation on these very values, the significant main effect of cortical network (F(1,103)=175.57) is at least partly a re-statement of the clustering criterion rather than an independent empirical finding. This compromises the abstract and discussion claims that hypothalamic subregions show stronger connectivity with Cortical Network 1 than with Cortical Network 2. The interaction and the subregion ordering are less directly forced, but the main effect should be re-derived with an independent definition of cortical networks (e.g., split-half derivation and test, a pre-specified network atlas, or a null model showing that k-means on noise does not produce such separation).
  2. [Methods, 'Hypothalamic' and 'Parcellations'; Results, 'Functional Subregions of the Hypothalamus'; Table 1] The four-community parcellation is supported only by consensus probabilities across 100 runs of Louvain on the same group-averaged matrix. This demonstrates algorithmic stability but not that the partition reflects neuronal organization: Louvain will partition any positive-semidefinite matrix, including one dominated by spatially smooth physiological noise or ventricular partial voluming. The kneedle-based exclusion of high-variability voxels removes extreme voxels but does not establish that the retained correlations are neuronal. The Louvain resolution parameter is not reported, so the number of communities is not reproducible. Please add a null-model comparison (e.g., surrogate data preserving the covariance or autocorrelation structure), split-half replication across subjects, or a formal concordance test against anatomical subfields, and report the resolution parameter.
  3. [Results, 'Functional Subregions of the Hypothalamus' and 'Cortical Connectivity with Hypothalamic Subclusters'] The full analysis chain—Louvain on the group-averaged matrix, k-means on the same subjects' data, and the ANOVA—uses the same participants for discovery and inference. Even if each step is locally stable, the reported p-values do not account for the selection of the parcellation and the network assignment on the same data. A split-half or leave-one-subject-out replication of both the hypothalamic communities and the cortical network contrast would substantially strengthen the generalizability of the claims and should be reported before the results are presented as a definitive functional architecture.
minor comments (5)
  1. [Figure 3 caption] The caption appears to swap the two network names: it says the hypothalamic communities showed greater connectivity with frontal and midline regions in Cortical Network 2 compared to posterior sensorimotor areas in Cortical Network 1, which contradicts the text and Figure 2, where Cortical Network 1 is the anterior/frontal/midline network.
  2. [Methods, 'Functional Connectivity Analysis'; Supplementary Figure 1] The text says the Calinski-Harabasz index was used to select the maximum index, while Supplementary Figure 1 describes the 'Kneedle' algorithm selecting an elbow; please clarify which criterion was used for the final two-cluster solution.
  3. [Methods, 'Parcellations' and 'Functional Connectivity Analysis'] The handling of run-level missing voxels is ambiguous: if a voxel is masked in one run but not another, pairwise correlations may be computed over different time-series lengths, and the text does not state whether concatenation, pairwise deletion, or imputation was used. Please clarify this and assess the potential impact on the group-level connectivity matrix.
  4. [Methods, 'Hypothalamic'] The choice of a 20% probability threshold for binarizing the Pauli hypothalamus mask and the use of the kneedle algorithm for high-variability voxel exclusion are not justified or validated; a sensitivity analysis over these thresholds would help establish that the four-community solution is not an artifact of these choices.
  5. [Acknowledgements] The sentence 'The views, opinions, and/or findings contained in this review are those of the authors' appears to use 'review' where 'manuscript' or 'paper' is intended.

Circularity Check

1 steps flagged · score 6.0 of 10

The two-network cortical coupling contrast is partially forced by k-means/ANOVA double-dipping.

  1. fitted input called prediction [Results – Functional Connectivity with Hypothalamic Subclusters (k-means and ANOVA paragraphs)]
    "To reduce multiple comparison concerns and simplify results, we performed a k-means cluster analysis to group together ROIs with similar connectivity profiles across hypothalamic subregions (i.e. across a 360 cortical ROIs x 4 hypothalamic subregion ROIs, matrix). ... We probed for differences between the cortical networks in their functional connectivity with the different hypothalamic communities using a 2 (cortical networks) x 4 (hypothalamic communities) repeated measures analysis of variance (ANOVA)."

    The two 'cortical networks' entered into the ANOVA are not independent of the connectivity values being tested. K-means was applied to the 360x4 matrix of Fisher z-values (cortical ROIs x hypothalamic subregions) to group ROIs with similar profiles, and the same z-values, averaged within the resulting clusters, are then compared in the 2x4 ANOVA. K-means partitions data to maximize between-cluster separation, so the significant main effect of cortical network (Cortical Network 1 > Cortical Network 2) is, at least in part, a restatement of the clustering objective rather than an independent confirmation of differential hypothalamic-cortical coupling. The p-values are double-dipped and inflated as evidence for the two-network architecture.

full rationale

The paper's four-community parcellation of the hypothalamus is not itself circular: it is generated from the within-hypothalamus 1019x1019 intrinsic connectivity matrix, and its comparison with the Makris et al. anatomical framework is an external anchor. The prior self-citations about 7T subcortical subregion identification are empirical feasibility evidence, not a unique theoretical premise that forces the present result. The main circularity is the k-means/ANOVA double-dipping: the cortical networks are fit to the same hypothalamic-cortical connectivity values that the ANOVA then presents as a significant differential-coupling finding. Because the clusters are constructed to maximize separation of these very connectivity profiles, the main effect of cortical network is partly manufactured by the clustering step. The paper also lacks a null model or split-half replication for the Louvain solution, which is a validation gap rather than a circularity, and there is an internal inconsistency between selecting the maximum Calinski-Harabasz index (Methods) and the kneedle elbow (Results). Overall, the headline two-network coupling contrast is partially forced, so the paper receives a 6.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The paper contributes a new data-driven parcellation, but it depends on several hand-set thresholds and standard neuroimaging assumptions. The most consequential choices are the 20% mask threshold, the per-run kneedle exclusion of high-variability voxels, the k=2 cortical cluster count selected on the same data used for inference, and an unreported Louvain resolution parameter. No invented physical entities are proposed.

free parameters (4)
  • Pauli hypothalamus mask probability threshold = 20%
    Threshold chosen by hand; determines the 1019 voxels entering all analyses.
  • High-variability voxel exclusion threshold = per-run kneedle value
    Kneedle sets a data-dependent cutoff on temporal variability to exclude third-ventricle voxels; different cutoffs alter the connectivity matrix.
  • Number of cortical k-means clusters = 2
    Selected by Calinski-Harabasz index with kneedle on the same data used in the ANOVA; k=2 defines the two compared cortical networks.
  • Louvain resolution parameter = not reported
    Community partitions (four communities) depend on the resolution parameter; absence blocks exact reproduction.
assumptions (5)
  • domain assumption Resting-state BOLD functional connectivity reflects underlying neuronal connectivity.
    The entire analysis interprets BOLD correlations as functional connections; physiological noise could contribute to the communities and cortical coupling values.
  • domain assumption The Pauli probabilistic hypothalamus mask, resliced and binarized at 20%, accurately delineates human hypothalamic tissue.
    Used in Methods 'Hypothalamic' to create the 1019-voxel mask; any error or threshold effect propagates to all subregions.
  • domain assumption Louvain community detection on the group-averaged functional connectivity matrix yields meaningful functional subregions.
    Community structure is algorithm- and resolution-dependent; no null model or cross-validation is reported.
  • domain assumption The Glasser cortical atlas remains valid after volumetric spatial normalization to MNI space.
    Cortical parcels are defined on surface anatomy; registration errors could misattribute hypothalamic coupling to wrong parcels.
  • ad hoc to paper The two cortical k-means clusters correspond to real cortical networks rather than artifacts of clustering the observed connectivity matrix.
    The clusters are derived from the same 360x4 data later submitted to ANOVA; labeling them as networks is an interpretive step specific to this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Functional Architecture of the Human Hypothalamus: Cortical Coupling and Subregional Organization Using 7-Tesla fMRI." pith.science (2026). https://pith.science/paper/CMCU2CET

@misc{pith2026250606191,
  author       = {Pith},
  title        = {Pith review of: Functional Architecture of the Human Hypothalamus: Cortical Coupling and Subregional Organization Using 7-Tesla fMRI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CMCU2CET}},
  note         = {Machine review of arXiv:2506.06191}
}
read the original abstract

The hypothalamus plays an important role in the regulation of the bodys metabolic state and behaviors related to survival. Despite its importance however, many questions exist regarding the intrinsic and extrinsic connections of the hypothalamus in humans, especially its relationship with the cortex. As a heterogeneous structure, it is possible that the hypothalamus is composed of different subregions, which have their own distinct relationships with the cortex. Previous work on functional connectivity in the human hypothalamus have either treated it as a unitary structure or relied on methodological approaches that are limited in modeling its intrinsic functional architecture. Here, we used resting state data from ultrahigh field 7 Tesla fMRI and a data driven analytical approach to identify functional subregions of the human hypothalamus. Our approach identified four functional hypothalamic subregions based on intrinsic functional connectivity, which in turn showed distinct patterns of functional connectivity with cortex. Overall, all hypothalamic subregions showed stronger connectivity with a cortical network, Cortical Network 1 composed primarily of frontal, midline, and limbic cortical areas and weaker connectivity with a second cortical network composed largely of posterior sensorimotor regions, Cortical Network 2. Of the hypothalamic subregions, the anterior hypothalamus showed the strongest connection to Cortical Network 1, while a more ventral subregion containing the anterior hypothalamus extending to the tuberal region showed the weakest connectivity. The findings support the use of ultrahigh field, high resolution imaging in providing a more incisive investigation of the human hypothalamus that respects its complex internal structure and extrinsic functional architecture.

Figures

Figures reproduced from arXiv: 2506.06191 by the authors.

Figure 1
Figure 1. The four hypothalamic subregions identified using community detection analysis and two cortical communities identified with k-means clustering analysis [PITH_FULL_IMAGE:figures/full_fig_p034_1.png] view at source ↗
Figure 4
Figure 4. Comparison of functionally derived subregions in this study and anatomical parcels made by Makris et al. 2013. 4a depicts the functional subregions identified in the present study in the medial view from the left hemisphere. and 4b depicts Makris et al.’s (2013; FIgure 2) parcellation of the hypothalamus based on anatomical landmarks, adapted from their paper. Note that the third ventricle is visible in the figure f… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

12 extracted references · 9 canonical work pages

  1. [1]

    J., Burns, C., Goncalves, M., Jarecka, D., Ziegler, E.,

    Esteban, O., Markiewicz, C. J., Burns, C., Goncalves, M., Jarecka, D., Ziegler, E., ... & Ghosh, S. (2020). nipy/nipype: 1.5. 1 [computer software]. Zenodo. Abraham, A., Pedregosa, F., Eickenberg, M., Gervais, P., Mueller, A., Kossaifi, J., Gramfort, A., Thirion, B., & Varoquaux, G. (2014). Machine learning for neuroimaging with scikit-learn. Frontiers in...

    work page 2020

  2. [2]

    Note that the third ventricle is visible in the figure from Makris et al

    parcellation of the hypothalamus based on anatomical landmarks, adapted from their paper. Note that the third ventricle is visible in the figure from Makris et al. (2013) in tan in. Ant. = Anterior, Pos. = Posterior , a-sHyp = ant erior-superior hypothalamus, a-iHyp = anterior-inferior hypothalamus, supTub = superior tuberal hypothalamus, infTub = inferio...

    work page 2013

  3. [6]

    A., Becerra, L., Johnson, A., Burstein, R., & Borsook, D

    https://doi.org/10.3390/ani11061733 Moulton, E. A., Becerra, L., Johnson, A., Burstein, R., & Borsook, D. (2014). Altered hypothalamic functional connectivity with autonomic circuits and the locus coeruleus in migraine. PLOS ONE, 9(4), e95508. https://doi.org/10.1371/journal.pone.0095508 Muschelli, J., Nebel, M. B., Caffo, B. S., Barber, A. D., Pekar, J. ...

    work page doi:10.3390/ani11061733 2014

  4. [8]

    E., Leak, R

    https://www.frontiersin.org/article/10.3389/fninf.2014.00014 Abrahamson, E. E., Leak, R. K., & Moore, R. Y. (2001). The suprachiasmatic nucleus projects to posterior hypothalamic arousal systems. NeuroReport, 12(2), 435–440. Andrews-Hanna, J. R., Reidler, J. S., Sepulcre, J., Poulin, R., & Buckner, R. L. (2010). Functional-anatomic fractionation of the br...

    arXiv 2001

  5. [10]

    B., & Stevenson, R

    https://www.frontiersin.org/article/10.3389/fnhum.2016.00244 Huey, R. B., & Stevenson, R. D. (1979). Integrating thermal physiology and ecology of ectotherms: A discussion of approaches. American Zoologist, 19(1), 357–366. https://doi.org/10.1093/icb/19.1.357 Johnson, R. F., Morin, L. P., & Moore, R. Y. (1988). Retinohypothalamic projections in the hamste...

    work page arXiv 1979

  6. [12]

    F., van der Kouwe, A., Abbs, B., Boriel, D., Handa, R

    https://doi.org/10.1186/s11689-017-9188-7 Makris, N., Swaab, D. F., van der Kouwe, A., Abbs, B., Boriel, D., Handa, R. J., Tobet, S., & Goldstein, J. M. (2013). Volumetric parcellation methodology of the human hypothalamus in neuroimaging: Normative data and sex differences. NeuroImage, 69, 1–10. https://doi.org/10.1016/j.neuroimage.2012.12.008 Margulies,...

    work page doi:10.1186/s11689-017-9188-7 2013

  7. [69]

    A., Bianciardi, M., Hartley, L., Matthewson, G., Choi, J.-K., Quigley, K

    Kragel, P. A., Bianciardi, M., Hartley, L., Matthewson, G., Choi, J.-K., Quigley, K. S., Wald, L. L., Wager, T. D., Barrett, L. F., & Satpute, A. B. (2019). Functional involvement of human periaqueductal gray and other midbrain nuclei in cognitive control. Journal of Neuroscience, 39(31), 6180–6189. https://doi.org/10.1523/JNEUROSCI.2043-18.2019 Kragel, P...

    work page doi:10.1523/jneurosci.2043-18.2019 2019

  8. [599]

    https://doi.org/10.1002/cne.901820402 Satopaa, V., Albrecht, J., Irwin, D., & Raghavan, B. (2011). Finding a “Kneedle” in a haystack: Detecting knee points in system behavior. 2011 31st International Conference on Distributed Computing Systems Workshops, 166–171. https://doi.org/10.1109/ICDCSW.2011.20 Satpute, A. B., & Lindquist, K. A. (2019). The default...

    work page arXiv 2011

Show all 12 references
  1. [940]

    https://doi.org/10.1038/s41597- 024-03775-2 Saper, C. B. (2012). Hypothalamus. In J. K. Mai & G. Paxinos (Eds.), The Human Nervous System (Third Edition) (pp. 548–583). Academic Press. https://doi.org/10.1016/B978-0-12-374236-0.10016-1 Saper, C. B., & Lowell, B. B. (2014). The...

    2012 doi

  2. [2013]

    and 4b depicts Makris et al.’s (2013; FIgure

    4a depicts the functional subregions identified in the present study in the medial view from the left hemisphere. and 4b depicts Makris et al.’s (2013; FIgure

    2013

  3. [6537]

    default mode

    Croizier, S., Amiot, C., Chen, X., Presse, F., Nahon, J.-L., Wu, J. Y., Fellmann, D., & Risold, P.-Y. (2011). Development of posterior hypothalamic neurons enlightens a switch in the prosencephalic basic plan. PLOS ONE, 6(12), e28574. https://doi.org/10.1371/journal.pone.00285...

    2011

  4. [6571]

    https://doi.org/10.1038/380243a0 Raichle, M. E. (2015). The brain’s default mode network. Annual Review of Neuroscience, 38, 433–447. Ranganath, C., & Ritchey, M. (2012). Two cortical systems for memory-guided behaviour. Nature Reviews Neuroscience, 13(10), 713–726. https://do...

    2015 doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.