Cyclic FISH Platform

Automating the rounds of a multiplexed FISH run

A multiplexed FISH experiment is the same protocol run again and again. Readout probes in, incubate, stringency wash, imaging buffer, image, then strip the signal and start the next bit of the codebook. Ten to twenty solution steps, fourteen to forty times over, and every one of them has to land on the same field of view as the last. By that point the biology is settled: the panel, the codebook, the number of bits. What decides whether the run finishes is the liquid handling underneath it. The routes open to you differ in how well they hold that up over days.

Manual pipetting. Where most runs start, and still the honest baseline. Every solution change puts a hand on the sample under the objective, so the flow cell can shift, the recorded field positions drift, and one bubble across the chamber can end the experiment. A fifty-step protocol spread over a weekend also means being there for it, including the 3 a.m. wash nobody wants to be awake for. Signal intensity varies with how each step was pipetted, and that is the part that hurts, because decoding a codebook depends on cycles being comparable.

A syringe pump with a rotary valve. The usual first build, and it does automate the sequence. Its ceiling is volume. A syringe carries a fixed amount. A multi-round run gets through litres of wash and imaging buffer. So someone comes back to refill it, in the middle of the run the automation was meant to cover. Each refill stops the flow and opens a way in for air. One bubble across the section can end an experiment that has already taken two days. At the low rates a section tolerates, the plunger advances in discrete steps. The seal and the tubing take up the first part of that movement. The flow then arrives in steps, and later than the setting suggests. Reagent left in the barrel at the end of a step is volume you do not recover, which stings most with a readout mix.

A peristaltic pump with a rotary valve. It answers the volume problem: it draws from a bottle for as long as the run lasts. And it brings its own. Rollers squeeze the tubing to move liquid, so the flow reaches the section as a train of pulses. The shear rises and falls with every roller that passes. Over forty rounds, that is forty opportunities to lift a section off its coverslip, and during an acquisition a pulse shifts the liquid in the field. The tubing is also a consumable under mechanical fatigue. Compressed thousands of times across a multi-day run, it drifts. The rate on day three sits off the one you calibrated on day one. Early and late rounds of a codebook then stop being comparable. In both builds, the positions on the rotary valve cap how many solutions the run can hold. Sequencing the pump, the valve and the microscope together also becomes a script you write and maintain yourself.

An integrated platform. Solutions are held on the platform and moved by air pressure, so nothing mechanical touches the liquid and no volume waits in a barrel. Thirty-four solutions and more sit ready, so the count grows with your codebook. The platform holds the flow rate you set, clears air out of the line, rinses between two solutions, and stops the flow for acquisition. Cycle one and cycle forty then arrive the same way. Your flow cell, your probes and your microscope stay yours.

Deliver every solution your codebook needs.

The round count is the hard constraint on this protocol, and it comes from the biology. A fourteen-bit barcode scheme needs fourteen readout hybridizations, HCR-based schemes have run to forty cycles, and each round needs its own probe mix on top of the buffers around it. The platform carries 34 solutions and more, so a sequence steps through the whole set before anyone opens the setup. A longer codebook is a matter of configuration, and we size it with you. Load the readout mixes and the buffers once, program the order, and the whole codebook is on the platform.

Hold the same exposure from the first cycle to the last.

Decoding a barcode depends on cycles being comparable, so what reaches the sample in cycle one should reach it the same way in cycle forty. That is a flow-rate question. You set the flow rate, and the platform holds it there, checking what actually flows and correcting itself as the run proceeds. The delivery stays the same from cycle one to cycle forty, so an incubation begins when you say it begins and a stringency wash arrives at the strength you set. Steady, pulseless delivery also keeps the wall shear stress on a tissue section low and predictable, which counts when that section has to come through forty rounds still attached to its coverslip.

Rule out cross-contamination between readouts.

A readout probe still sitting in the line when the next one arrives shows up as signal in the wrong bit. That is cross-contamination, and afterwards there is no way to separate it from biology. The platform rinses the line to your sample between two solutions of interest, so every probe mix travels a clean line on its way in. You decide where a rinse belongs: between each readout hybridization, or only at the steps your chemistry cares about. It runs as a programmed step like any other.

Load an expensive probe mix in a small volume.

On the platform, the reservoir’s volume has no bearing on the injection. A 2 mL tube of readout mix injects at the same flow rate as a litre of wash buffer, and swapping one for the other changes nothing in the settings. The two ends of this protocol sit far apart: probe mixes are the expensive reagent and belong in small aliquots close to the sample, while stringency and imaging buffers go through in volume across a multi-day run. Both run on the same platform, on the reservoirs you choose. You set the flow rate for a step, then choose what ends it: the volume you want injected, or how long the injection lasts. A readout step ended by volume stops once that volume has passed, so an expensive mix is spent on the sample and nowhere else.

Start a run without a laptop at the microscope.

A front-panel touchscreen sets the flow rates and builds the sequence, so a run starts with no computer to connect first. In a dark imaging room, where a bright screen is unwelcome next to a sensitive detector, that is worth something. Build the sequence once, store it on the platform, and recall it for the next sample, or load a pre-defined one. The order, the volumes and the incubation times then repeat exactly, whoever starts the run.

Protect the sample, and the next one.

A section can represent weeks of work, and sometimes clinical material that cannot be replaced, so committing it to a run of several days means trusting the fluid path. The platform makes that check itself. Every run opens with an initialization step it performs on its own, at every use. Nothing for you to launch, nothing to inspect. The setup is verified before anything reaches your sample. Air is the other thing that ends a run of this length. The platform removes bubbles from the line on its own. A bubble that forms overnight is dealt with when it forms. The run then primes the lines. The first solution reaches your sample through a filled path, on the same terms as every solution after it. From there the sequence you built or loaded proceeds on its own. It closes with a cleaning step that clears the reagents out, so the next experiment starts on a clean platform. Everything the run did comes out as data you can export.

Keep imaging and solution changes on one clock.

A cycle only works if the two halves agree on the timing: the image is taken once the imaging buffer is in and the flow has stopped, and the next solution starts moving once the acquisition is finished. A trigger line carries that. The platform signals the microscope to acquire at the imaging step, and holds the sequence while the acquisition runs, so a fourteen-cycle experiment proceeds as one sequence, with no one timing each step by hand. Where a protocol needs the liquid completely still for acquisition, the platform can close the line at the imaging step, so nothing drifts in the field while the camera integrates.

Reproducible multiplexed imaging.

Most spatial experiments fail not only on the biology, also on the setup: a wash that ran short, a probe mix that reached the sample at a different rate, a cycle someone did at the end of a long day. A stored sequence and a held flow rate take that variable out. The protocol expertise stays yours; what changes is that the same protocol delivers the same exposure on Tuesday and on Saturday, so a reviewer’s question about reproducibility has an answer in the run record.

A traceable record for regulated work.

As a sequence proceeds, the platform logs its full history, every step it ran and every value it held, with a timestamp. That log is a 21 CFR Part 11-compliant record, meeting the FDA rule for electronic records and audit trails, so what reached your sample is captured and holds up to review. For a pharmaceutical lab or a facility running work for others, the reagent exposure becomes documented, auditable data, ready for a validation file.