How scientists made pig and human brain maps comparable

A pig brain cannot simply be shrunk, rotated and placed beside a human brain to see whether the two work alike. Their shapes and scales differ. They were scanned under different conditions and mapped against different anatomical references. Even the two MRI methods used in a new comparison measure different physical signals.

That difficulty is precisely what makes the comparison useful. If researchers can identify which similarities survive the translation between species—and which depend on an analyst’s choices—they gain a better basis for judging animal models of the human brain.

In a peer-reviewed paper published September 22, researchers compared large-scale activity patterns and estimated white-matter pathways in pigs and humans. They reported correspondences in several familiar network systems, including sensorimotor, visual, frontal, cerebellar, default-mode and executive patterns. The result is not that the two brains are interchangeable. It is a worked example of how scientists compare two decidedly unlike objects without pretending that resemblance settles every question.

Two kinds of map

The study assembled resting-state functional MRI data from 42 pigs and 100 unrelated human participants. For structural comparisons, it also used diffusion MRI from 30 participants in the Human Connectome Project, according to the study’s full methods and results.

Both species’ functional scans came from 3-tesla MRI systems, and both recorded an image every three seconds. Matching that timing removed one obvious source of disagreement. Other differences remained: the pigs were scanned under isoflurane anesthesia, while the humans were awake, and the species required different coils, image resolutions and anatomical reference maps.

The researchers then built two fundamentally different kinds of map:

• Resting-state functional MRI followed slow changes in blood oxygenation while no specific task was being performed. Regions whose signals rose and fell together were treated as members of a possible functional network.

• Diffusion MRI measured how readily water moved in different directions through tissue. A tractography algorithm used those directions to reconstruct plausible paths through white matter, the tissue containing long neuronal fibers.

Neither method watches thoughts or traces individual neurons. Blood-oxygen-level-dependent, or BOLD, fMRI is a vascular signal associated with neural activity. Classic simultaneous recordings showed that it was especially related to local field potentials—summed electrical activity around a recording site—rather than simply mirroring neurons’ output spikes (Nature). Diffusion tractography goes one step further into inference: it turns local estimates of water direction into continuous streamlines that may represent fiber bundles.

That distinction matters. The functional maps describe correlated signals; the structural maps estimate possible routes. Agreement between them can strengthen a case, but neither is a direct photograph of information moving through a brain.

Finding networks without assuming their locations

For the functional comparison, the team used group independent component analysis, or ICA. In plain language, ICA attempts to separate a complicated mixture of signals into recurring patterns whose changes are statistically distinguishable from one another.

The researchers ran the analysis separately for pigs and humans, initially producing 70 components in each species. They then selected 12 components using their spatial and temporal characteristics and visual inspection, grouping those components into seven named networks: frontal, sensorimotor, visual, default mode, cerebellar, auditory and central executive.

This approach has an advantage: it does not require a pig region to occupy the same coordinates or possess the same shape as its proposed human counterpart. Each species gets its own decomposition first. The comparison comes afterward.

It also introduces judgment. An algorithm generated the components, but people decided which ones looked biologically meaningful and how they should be paired and named. “Independent component” is therefore closer to the direct analytical result than “pig default-mode network,” which is an interpretation based on location, signal behavior and comparison with existing atlases.

The clearest correspondences involved patterns with relatively recognizable geography. Sensorimotor components covered regions associated with movement and bodily sensation in both species. Visual components occupied corresponding posterior visual territory. Frontal and cerebellar patterns also showed spatial and temporal similarities. The authors additionally matched default-mode and central-executive components, although those higher-order labels depend more strongly on how regions and networks are translated between species.

Across the selected components, the authors reported that corresponding networks tended to resemble one another more than nonmatching networks. Their temporal analysis particularly highlighted frontal, sensorimotor, visual and cerebellar correspondences. This is evidence for shared large-scale organization—not evidence that the networks support identical experiences or cognitive abilities.

A broader review of cross-species fMRI describes the same central problem: mammals can share conserved network principles while also possessing species-specific cortical arrangements. Researchers increasingly compare whole networks, connectivity patterns and broad gradients rather than demanding a brittle one-region-to-one-region match.

The anesthesia-shaped uncertainty

One mismatch cannot be processed away: the pigs were anesthetized and the humans were awake.

Anesthesia does not necessarily erase the spatial outline of every network. A study that scanned marmosets both awake and under isoflurane found that much of the networks’ spatial structure remained recognizable. But isoflurane generally weakened connectivity and substantially altered some interhemispheric and thalamic relationships (Cerebral Cortex).

That result offers two cautions at once. A network visible under anesthesia need not be an artifact, but its measured strength and relationships may differ from those of an awake brain. In the pig–human study, anesthesia could therefore obscure genuine correspondence, create an apparent mismatch or change which components look easiest to align.

Other processing choices matter too. Pig functional images had lower temporal signal-to-noise than the human data, meaning the desired fluctuations were less distinct from unwanted variation. Mapping a folded pig cortex onto a surface reference was also less precise than the corresponding human registration. Those limitations do not nullify the observed patterns; they narrow what can safely be concluded from them.

Streamlines are hypotheses, not axons

The structural half of the study used diffusion tensors and deterministic tractography to estimate major white-matter pathways. Color-coded maps indicated the dominant direction of water diffusion, while reconstructed streamlines supplied a visually intuitive account of pathway geometry.

The authors found substantial similarities in the arrangement of major pathways. This supports correspondence in broad white-matter geometry: both brains contain large bundles traveling through comparable relative positions and directions.

Yet tractography’s polished curves can look more certain than they are. Fibers cross, bend and merge within image voxels, leaving multiple possible routes consistent with the same diffusion measurements. An international ground-truth tractography challenge found that leading algorithms recovered much of the simulated anatomy but also generated many invalid bundles, some repeatedly across research groups.

So a reconstructed line means “the diffusion data and model permit this route,” not “the scanner observed an axon following this route.” Differences in voxel size, diffusion directions and acquisition protocols between the pig and human datasets add another reason to treat fine-grained comparisons cautiously.

A four-question test for persuasive brain maps

The study suggests a reusable way to read almost any cross-species imaging claim. When two colorful maps appear strikingly alike, ask:

1. What physical quantity was measured? Here it was blood oxygenation or directional water diffusion—not cognition and not individual neural connections.
2. What did software construct from that measurement? ICA produced candidate activity patterns; tractography produced possible white-matter paths.
3. Where did human judgment enter? Researchers selected 12 of 70 components, paired them across species and attached network names.
4. Were the scanning conditions comparable? Timing and magnetic-field strength were matched in useful ways, but wakefulness, image quality, anatomy and some acquisition details were not.

Passing those questions does not require a perfect experiment. It requires conclusions proportionate to the evidence.

The direct observations here are MRI signals from two species. The analysis found recurring components and estimated pathway geometry. The supported inference is that pig and human brains share parts of their large-scale functional and structural organization. The study does not demonstrate equivalent cognition, make the brains interchangeable or establish that pigs are suitable models for every neurological condition.

That boundary is not a disappointment. It is the interesting achievement: researchers found recognizable common organization across two differently shaped brains while leaving visible the chain of measurements and choices that made the resemblance possible.