NeuroTrailblazers banner

H01 Human Cortex Fragment

Overview

The H01 dataset represents a watershed moment in neuroscience: the first nanoscale reconstruction of human brain tissue at synaptic resolution. Published by Shapson-Coe et al. in Science (2024), this dataset derives from a small fragment of temporal lobe cortex — approximately 1 mm³ — removed during surgical treatment of drug-resistant epilepsy in a 45-year-old woman. Imaged at 4 nm XY and 33 nm Z resolution, the resulting dataset comprises roughly 1.4 petabytes of imaging data and contains approximately 57,000 cells, 150 million synapses, and about 8,000 neurons with cell-body-containing profiles.

For decades, our understanding of cortical circuitry has been built primarily on rodent models. H01 provides the first opportunity to examine human cortical wiring at the level of individual synapses and to ask: how much of what we have learned from mice actually applies to humans?

The Source Tissue: Surgical Resection

Clinical Context

The tissue was obtained from a surgical resection performed to treat medically refractory temporal lobe epilepsy. In such procedures, a portion of the temporal lobe (including the epileptic focus) is removed to reduce seizure frequency. The resected tissue, which would otherwise be discarded, was redirected for research with informed consent.

This clinical origin has profound implications for the dataset:

Tissue Characteristics

Human cortical tissue differs from rodent tissue in several ways that are immediately apparent in the EM data:

Technical Pipeline

Acquisition

The tissue block was sectioned using automated ultramicrotomy and imaged using multi-beam scanning electron microscopy, similar in principle to the MICrONS pipeline but adapted for the specific challenges of human tissue. The 33 nm section thickness (slightly thinner than the 40 nm used in many mouse studies) was chosen to improve z-resolution and aid in tracing fine neuronal processes through the volume.

Segmentation and Reconstruction

Automated segmentation was performed using deep learning models trained on human tissue ground truth. The segmentation pipeline had to contend with features not present in rodent training data, including lipofuscin granules, larger cell bodies, and the fixation quality gradient. Additional post-processing steps were developed to handle these human-specific challenges.

Scale of the Dataset

At 1.4 petabytes, H01 is comparable in raw data volume to MICrONS. The reconstruction identified approximately:

These numbers are approximate because the boundaries of the well-preserved tissue are not sharp, and inclusion criteria for “reconstructed” neurons depend on the analysis.

Key Scientific Findings

Discovery of Axon Whorls

Perhaps the most unexpected finding in H01 was the discovery of “axon whorls” — unusual, tightly wound tangles of axonal processes that had not been previously described in the neuroanatomical literature. These structures consist of one or more axons forming complex loops and spirals, sometimes encapsulating other cellular elements.

The significance of axon whorls remains unclear. Hypotheses include:

The existence of axon whorls illustrates a key principle: when you image tissue at a resolution never before achieved, you will find things that nobody predicted. The connectomics community must develop frameworks for interpreting novel structures that do not map onto existing knowledge.

Neurons with Unusually High Synapse Counts

A small population of neurons in H01 was found to have synapse counts far exceeding the expected range. These “hyper-connected” neurons received or made many more synapses than their neighbors, raising questions about whether they represent a distinct functional class, a pathological feature, or a normal extreme of the connectivity distribution.

Layer-Specific Connectivity

Despite the tissue being from a single cortical region and a single individual, H01 revealed clear layer-specific differences in connectivity density, cell type composition, and synapse ultrastructure. These laminar patterns are broadly consistent with what is known from rodent studies, suggesting that the basic laminar organization of cortex is conserved across mammals.

Key layer-specific findings include:

Comparison with Mouse Cortex

Direct comparison between H01 and mouse cortical connectomics datasets (particularly MICrONS) reveals:

Data Access

The H01 dataset is publicly available for browsing and analysis:

Unlike FlyWire and MICrONS, H01 does not currently use the CAVE infrastructure, which means that collaborative proofreading and annotation versioning are handled differently. This reflects the dataset’s origin as a collaboration led by Google Research and the Lichtman lab at Harvard.

Challenges and Interpretive Cautions

Fixation Quality

The immersion fixation protocol results in a gradient of tissue preservation across the volume. Regions near the tissue surface are well preserved, with clear membrane contrast and identifiable synaptic specializations. Regions deeper in the block may show extraction artifacts, membrane disruption, or reduced contrast. Analyses must either restrict themselves to well-preserved regions or explicitly account for preservation quality as a covariate.

Pathological Features

Because the tissue was resected from an epileptic brain, some features in the dataset may reflect pathology rather than normal anatomy. Reactive gliosis, neuronal loss, aberrant sprouting, and other epilepsy-related changes could be present. The challenge is that we do not have a “normal” human connectomics dataset for comparison — H01 is the only dataset of its kind.

Generalizability

A single 1 mm³ fragment from one person’s temporal lobe cannot represent the full diversity of human cortical architecture. Cortical regions differ in their laminar structure, cell type composition, and connectivity patterns. Age, sex, genetic background, and life experience all influence brain structure. H01 is a proof of concept, not a definitive atlas.

Ethical Considerations

Human tissue connectomics raises ethical questions that do not arise with animal tissue:

Significance for the Field

Human Connectomics Is Feasible

H01 demonstrates that the technical pipeline developed for rodent connectomics — automated sectioning, multi-beam SEM, deep learning segmentation — can be applied to human tissue. The pipeline requires adaptation (handling lipofuscin, larger cells, fixation gradients) but is fundamentally workable.

A Reference for Cross-Species Comparison

As more species are mapped at synaptic resolution (fly, worm, mouse, and now human), a comparative connectomics framework is emerging. H01 anchors the human end of this spectrum and enables direct comparison of circuit motifs across species.

A Foundation for Clinical Connectomics

In the long term, nanoscale connectomics of human tissue could contribute to understanding neurological and psychiatric disorders. If wiring differences can be linked to disease states, connectomics could complement genomic and functional approaches to diagnosis and treatment. H01 is the first step on this path.

Discussion Questions for Instructors

  1. How should researchers handle the interpretive challenges posed by pathological tissue? What controls or comparisons would strengthen conclusions drawn from H01?
  2. Lipofuscin granules are abundant in human neurons but rare in young rodent tissue. What challenges do they pose for automated segmentation, and how might training data need to be adapted?
  3. The discovery of axon whorls was entirely unexpected. How should the field approach novel structures that do not map onto existing anatomical knowledge?
  4. Compare the ethical considerations of human tissue connectomics with those of animal tissue connectomics. Are the ethical frameworks fundamentally different?
  5. If you were designing the next human connectomics project, what tissue source, brain region, and clinical context would you choose, and why?

Key References