Imaging individual neurons deep in the brain requires implanting a gradient-index (GRIN) rod, but ultrathin (0.35–0.5 mm) rods have intrinsic optical aberrations that keep only the centre of the field of view sharp. The authors designed aspheric corrective lenses by optical simulation and fabricated them with two-photon lithography (3D micro-printing), then replicated them onto the coverslip above the rod, so the probe needs no change to a standard two-photon microscope. Simulations predicted a 2–9× larger diffraction-limited field of view; bead measurements showed a 3.2–9.4× larger effective

eLife 2020 · Antonini, Sattin et al.

A lens thinner than a hair, 3D-printed onto a needle-thin brain probe, sharpens the edge of its view.

Simulated neurons: each glowing cell body (dark nucleus) sends out branching dendrites, and flashes when it fires. How sharp the view is at each distance from the centre follows the paper’s measured axial resolution of the Type II probe, without and with its lens (Figure 3e). The dashed ring marks where the focus is still shorter than 10 µm; the close-up magnifies the spot under the crosshair.

Deep in a mouse’s brain, a cluster of neurons relays every twitch of the whiskers to the cortex. To watch hundreds of them at once, you look down a glass rod as thin as pencil lead — and only the middle of the view is sharp.

These researchers 3D-printed a lens thinner than a hair onto that rod. Try it.

1Press Add printed lens and watch the edge of the view.

2Tap anywhere in the circle to magnify the neurons there. Near the edge, can you still see their branches?

01 · Why look through a needle?

The brain’s messages are written by crowds

Neuroscientists increasingly think the brain’s messages are carried not by single neurons but by groups: which cells fire together, and how tightly, is itself information. Reading such a population code means recording many individual cells at once, sharply, and without inventing relationships between them.

A two-photon microscope can do that near the surface. Its infrared laser makes a glowing protein shine only at the tiny point where the light is most concentrated, so cells can be picked out one at a time. But brain tissue scatters light like fog: beyond about 1–1.6 mm the focus falls apart. The cells studied here, in the thalamus’s whisker relay (the VPM ), sit about 3 mm down.

So you need a way in. Pick one:

Tissue column: 0.5 mm wide

20 of 20 fibres still connected

signals from the VPM to the whisker cortex

feedback from the cortex to the VPM

A 0.5 mm probe slides down a narrow track to just above the VPM. Little tissue is displaced, so the long fibres linking thalamus and cortex can stay intact.

Does the thin probe really spare the wiring?

The VPM is constantly talked back to by the cortex, so cutting those connections would change what you are trying to measure. The team labelled both directions of traffic and compared mice with and without the 0.5 mm probe. The labelled area in the whisker cortex was not significantly different (green ≈17% vs ≈15%; red under 2% in both; three mice each, measured).

The thin probe solves the access problem. But it brings an optical one.

02 · A rod that is a lens

Light that weaves, and rays that miss

The probe is a GRIN rod (gradient-index): glass whose refractive index — how strongly it slows light — is highest along its axis and falls toward its wall. Light inside keeps bending back toward the middle, so rays weave down the rod in long waves and come together again. The rod is a lens stretched into a needle.

Thin rods are imperfect. Rays leaving one point should meet again at one point, but they miss slightly — an aberration . Near the centre of the view the miss is tiny. Near the edge, rays cross at different depths and the focus smears into a long spindle.

The fix is a cancellation : bend the light a little the opposite way before it enters the rod, so the rod’s error undoes yours. That is the printed lens’s job.

Slide the scan spot toward the edge and watch the focus smear.

Turn up the correction strength.

Find the strength that keeps the view sharp furthest out.

Focus length here: 9.1 µm · sharp needs < 10 µm

Sharp out to 50 µm from the centre (of 250)

No lens: only the centre of the view is sharp.

teaching model At 0% and 100% the focus length follows the measured axial resolution of the Type II probe (Figure 3e); other strengths are interpolated for illustration. Ray paths are schematic and the rod’s width is exaggerated. The real lenses were designed with 3-D optical simulations at the laser’s 920 nm wavelength.

In simulations the team used the Strehl ratio : the peak brightness of the real focus divided by that of a perfect one. Above 0.8 (the Maréchal criterion) a focus counts as “diffraction-limited”, as good as optics allow. For the Type I probe, simulations predicted this out to ~165 µm from the centre with the lens versus ~70 µm without, about 5× the area; across the four designs, 2–9×.

In measurements they used the length of the focus along the light’s path (the axial resolution, measured as the full width at half maximum of a 0.1 µm bead’s image). Where it stayed below 10 µm — shorter than a neuron — the view counted as “effective”. That is the ring in the opening view.

03 · Printing the lens

A lens you could lose on a fingertip

The right correction is not a simple dome. The team’s optical simulations, which started from the classic shape used to correct telescope mirrors (a Schmidt plate), produced an aspheric profile: a shallow bowl with a raised rim, 0.5 mm across and less than 40 µm high — thinner than a human hair. Nothing that small and that oddly shaped can be ground and polished.

So they printed it. In two-photon lithography , a tightly focused laser hardens a light-sensitive resin only at its focal point, where the light is intense enough. Steering the focus through a cloud of points about 0.3 µm apart writes the solid shape layer by layer. The printed master was then copied by moulding — a silicone negative filled with UV-curing optical glue — straight onto the glass coverslip that sits on the rod. Nearly every lens came out usable.

a human hair, ≈70 µm

the printed lens: 500 µm wide, under 40 µm tall

the probe below it: 500 µm wide, like 0.5 mm pencil lead

04 · When one pixel hears two cells

How a blurry edge invents friendships

Why does a smeared focus matter beyond looking soft? The image is built point by point: at each pixel the laser focus sits in the tissue and collects the glow from everything inside it. A neuron is about 15 µm across. A 9 µm focus hears one cell. A 25–50 µm focus, at the edge of an uncorrected probe, can also hear a cell above or below.

The cells make GCaMP , a protein that glows brighter each time the cell fires. So hearing another cell means its flashes leak into your recording. Now picture two neighbours whose recordings both catch the same cell stacked between them…

Predict first. In a tissue where nearby cells are truly in sync by the same amount everywhere, what will the recorded sync of neighbours do toward the edge of an uncorrected probe?

Cells leaking into this recording: 0 · match with the target cell: r = 0.00

Press record to sample neighbour pairs across the view.

teaching simulation In the spirit of the paper’s synthetic calcium data (Figure 4a): pairs of neighbouring cells with a fixed true amount of shared input, plus 1–3 cells stacked above or below; focus length from the measured Type II curves. The size of the effect depends on how crowded the tissue is.

The team filled a 500×500×80 µm block with simulated neurons (spheres ~8 µm in radius with a dark nucleus, ~83,000 cells per mm³), gave them random spikes plus shared inputs, turned spikes into GCaMP-like glow, and “imaged” the block by sweeping focus shapes copied from the measured ones along the probes’ curved focal surfaces. Because the true activity of every cell was known, they could check how faithfully each probe recorded it. Without the lens, nearby-cell correlation rose with distance from the centre (slope significantly positive); with it, the slope was not different from zero. Corrected traces also matched each cell’s true activity better, picked up less from neighbours, and had higher signal-to-noise (mean 4.0 vs 2.3).

05 · Lens on, lens off

The test on real samples

Simulations are predictions. Here are the same probes with and without the printed lens, on real samples. Drag the divider.

Measuring the focus itself

Beads 0.1 µm across are far smaller than the focus, so each bead’s image is the focus. The team imaged beads at increasing distances from the centre and measured the focus length. Where it stayed under 10 µm (dotted line), they called the view effective. Choose a probe design:

Dashed red: no lens; solid blue: printed lens; shaded: the effective view. Source data for Figure 3 and Supplementary Table 3; mean ± s.d. from at least three probes per type. measured

06 · In awake mice

Cleaner signals, honest sync

Finally, living tissue. The team imaged GCaMP-labelled VPM neurons in awake mice through Type II probes, recording with the printed lens and after removing it. The two key predictions from the simulations held up.

Signal-to-noise

Mean SNR of 557 vs 306 recorded cells; p = 0.0011. The number of cells found per area was not different: the gain is quality, not quantity. measured

Sync of nearby cells vs distance from centre

Without the lens, sync crept up toward the edge: a small but significant rise (p = 0.006, permutation test). With the lens the trend was not significantly different from zero — as the simulations predicted. Dataset 2 was recorded during the behaviour experiments below. Redrawn from Figure 4m source data. measured

07 · What the wider window revealed

Listening to the whisker relay

Mice explore with their whiskers, and the VPM passes that touch to the cortex while receiving strong feedback from it. With the corrected probes the team recorded VPM cells while the mice sat quietly, whisked, or ran on a wheel, with cameras tracking whiskers and pupil size.

illustration Cells and whisking are simulated: ~90% of co-active groups brighten when whisking, ~10% dim, as the paper found.

0.19 bits ± 0.03

How much the population’s activity tells you about whether the mouse is whisking (1 bit would be a perfect yes/no). Mean ± s.e.m. over 24 recordings from four mice; Figure 7e source data. measured

Active brains are louder. When the mice whisked and ran, VPM cells fired more often (1.95 → 2.22 events per second on average) and more strongly (ΔF/F 0.137 → 0.245).

Information is sparse and cell-specific. Many cells carried a little information about whisking, a few carried a lot — and those few were often surrounded by low-information neighbours.

The edge is as trustworthy as the centre. A single cell’s information did not depend on where it sat in the view; yet adding cells from further out kept adding information the central cells did not already carry (slider above).

Co-active groups span the whole view. “Modules” of cells that fire together, found with a pattern-finding method (non-negative matrix factorization), stretched across hundreds of micrometres. Most (89.6%) were boosted by whisking; about 1 in 10 (10.4%) were suppressed. A cell could belong to several modules with a different weight in each.

08 · Back to the needle

A wider eye, same thin needle

We began with a question the authors care about: how do crowds of neurons deep in the brain share the work of representing the world? Answering it needs reach, sharpness across a wide view, and measurements that do not invent relationships between cells. The printed lens adds the last two to a probe that already had the first — without making it any thicker.

What changes: a ready-to-use probe that fits standard two-photon microscopes; sharpness even enough that cells at the edge can be compared fairly with cells at the centre; and a design recipe the authors expect to work for rods of other lengths and widths, to reach other deep regions.

What the authors see next: correcting longer rods; lenses for imaging whole 3-D volumes (their simulations already show a 1.7–3.7× larger sharp volume); pairing with holographic light patterns that switch chosen neurons on or off (optogenetics); and minimally invasive imaging for tissue diagnostics.

One more look at the edge:

Source and credits

Antonini A, Sattin A, Moroni M, Bovetti S, Moretti C, Succol F, Forli A, Vecchia D, Rajamanickam VP, Bertoncini A, Panzeri S, Liberale C, Fellin T (2020). Extended field-of-view ultrathin microendoscopes for high-resolution two-photon imaging with minimal invasiveness. eLife 9:e58882. doi:10.7554/eLife.58882

Figures and source data are from the article above, published under a Creative Commons Attribution 4.0 licence. Used here: Figure 3f; Figure 3—figure supplements 1b and 5b; Figure 5b,c; Figure 6b; Figure 7j. Images were cropped and resized; some labels and scale bars were redrawn. The animated brain-slice sketch is a simplified redrawing after Figure 3—figure supplement 7e. Charts were redrawn from the source data for Figures 3, 4 and 7 and from Supplementary Table 3.

Interactive views are simplified teaching simulations built on the paper’s measured curves; every measured value is labelled as such. Simulated neurons, their activity, the brain-slice sketch with its fibres and signals, the ray paths and the printing animation are illustrations.