Photothermal Recycling Biosensing
for Longitudinal Biochemical Monitoring

ASME IMECE 2026, November 9, 2026

Jing Pan, Assistant Professor
Yongchen Tai, Yunshen Li, Yida Wang, Jacob Lee
Mechanical and Aerospace Engineering, University of Florida

---

title - opening

Good morning. I’m Jing Pan, from Mechanical and Aerospace Engineering at the University of Florida, and this is work from my group led by Yongchen Tai.

The question behind it is simple to state: can we measure a low-abundance biomarker continuously, the way we already measure glucose? The answer we have been building is a thermal one, which is why I think it belongs at this meeting.

Longitudinal Biochemical Monitoring

?

On-Board Diagnostics (OBD)

Continuous Biochemical Monitor

c01 - a car reports its state continuously

A modern engine carries a few dozen sensors and reports its own state continuously. When something drifts, you know before it fails.

A body has far more state variables and worse consequences, and we sample it a few times a year, by drawing blood and sending it to a lab.

That gap is what this talk is about: a continuous measurement of the chemistry.

Example: Diabetes Management

• over 97,000 hypoglycemia in US annually
• insulin absorption rates differ among individuals
• insulin test: blood draw and lab analysis
• continuous insulin monitoring is not available

c02c - one example: the diabetes control loop

The clearest example of what biochemical dynamics are worth is diabetes.

Glucose is now measured continuously. That one measurement closes a control loop - a sensor on the body, an algorithm, an insulin pump - and the artificial pancreas changed how the disease is managed.

Now look at what the loop actually knows. It sees glucose. Insulin, the variable it is dosing, is invisible to it. On the left are two patients from the same closed-loop trial: insulin peaks about an hour after a dose in one, and three hours in the other. The controller runs on a model of that, and the only way to check the model is a blood draw and a lab.

What this loop is missing is an insulin sensor. Glucose sits at millimolar; insulin at picomolar, eight orders of magnitude lower. That gap is the technical problem for the rest of this talk.

Biomarker Measurement and Monitoring

Clinical/Bioanalytical Assays

Wearables

• In-patient care
• Accurate and comprehensive
• Invasive
• Slow, Expensive
• Sparse
• Limited to lab infrastructure

• At-home or wearable
• Limited analytes
• Non-Invasive
• Fast, minimal cost per sample
• Continuous, On-demand
• Limited specificity

Need: continuous and sensitive measurement of state-specific biomarkers

c03 - two regimes

Today you get one of two things.

The clinical laboratory is accurate and comprehensive, and it is invasive, slow, expensive and sparse. Wearables are continuous and cheap per sample, and they measure physical variables - heart rate, oxygen, temperature - plus a very short list of chemistries.

The intersection, continuous and sensitive measurement of a state-specific biomarker, is empty. That is the target.

Current Continuous Assays and Technical Challenges

Continuous aptamer sensors existin vivo, for hours, at micromolar

What the clinic needsdynamic, low-abundance biomarkers

• Clinically informative proteins, hormones and immune mediators sit at pM to fM
• They are dynamic: the trajectory carries the information
• Technical challenge: longitudinal monitoring of low-abundance biomarkers

c04b - current continuous assays and the low-abundance challenge

I should be careful here, because continuous biochemical sensors do exist and some of them are very good.

Aptamer sensors have run in vivo, in animals, for hours - the microneedle device on the left tracks two drugs at once, and the trace below follows an antibiotic through a rat for three hours. They work.

But look at the concentration axis. These are micromolar drugs, and the signal is a fractional change on a large background. The proteins, hormones and immune mediators that carry clinical information - the ones changing in that heatmap - sit at picomolar and below. That is six orders of magnitude, and it is where every continuous platform stops today.

The challenge is longitudinal monitoring of low-abundance biomarkers. And the reason it is hard is thermodynamic - that is the next slide.

Photothermal Recycling (PTR) Biosensing

c05 - photothermal recycling

Here is the problem stated properly.

A sensor that sees a picomolar analyte needs a binder with a very low dissociation constant. A binder with a very low dissociation constant does not let go. So the sensor saturates, and after the first measurement it is a record of the past, not a reading of the present.

Strong binding buys sensitivity and costs reversibility. You cannot fix that by rewriting the molecule - it is one equilibrium constant serving two requirements.

So we drive it instead. The surface is a thin gold film. A brief pulse of light heats it locally by about twenty degrees - enough to melt the complex off - and the surrounding fluid cools it straight back. The binder is not weakened; it is reset.

Measure, heat, cool, measure again. That is photothermal recycling: the thermal cycle of PCR, borrowed for sensing, with light as the heater and the buffer as the heat sink.

PTR Proof-of-Concept Demonstration

Assay Design and Optical Instrumentation

c06 - proof of concept

The assay is a surface-tethered DNA hairpin on a gold film in a flow cell. The 405-nanometre heating beam and the 488 imaging path share the objective, and a beam shaper makes the heating uniform across the field.

On the right is what a cycle looks like. Signal rises as the complementary strand binds, the heating pulse fires, the signal drops to baseline, and the next cycle starts from the same place. Cycle after cycle, no drift.

That is the control that says the mechanism itself works.

PTR Proof-of-Concept Demonstration

Assay Design and Optical Instrumentation

c07 - proof of concept, the reversible part

Same assay and instrument, with the concentration stepped.

The raw heating and cooling cycles on the left; on the right the PTR trace following the commanded concentration, and the control at fixed concentration staying flat.

Photothermal Effect on Au Film Modeling and Measurement

Prof. Qiye Zheng
HKUST

c13 - how hot the surface actually gets

Everything rests on one physical claim: a short pulse of light on a gold film raises the local temperature enough to release what is bound, without cooking the rest of the assay.

So we modelled it and then measured it. The model is finite-element heat transfer in COMSOL, with a localised surface heat-generation boundary condition at the illuminated spot. The measurement uses the temperature coefficient of resistance of the film itself - the film is its own thermometer.

At 405 nanometres and 13.6 watts per square centimetre the film comes up about twenty-one kelvin, repeatably, cycle after cycle, and the model agrees - 21.9 for a uniform source, 21.0 for a Gaussian one.

About twenty-one degrees is the number the whole platform is built on. It is enough to open a DNA duplex or an aptamer complex, and modest enough that the surface chemistry survives repeated cycling. This part is with Qiye Zheng, who brought the thermal transport side.

Improving Assay Sensitivity with Digital Counting

SIMOA Assay (<fM LOD)

c08 - counting instead of averaging

The second ingredient is how we read it.

Averaging intensity over a surface throws away the fact that binding is discrete. If instead you resolve single binding events and count them, the noise floor is set by counting statistics rather than by background brightness. That is the digital assay idea - it is what took ELISA to sub-femtomolar.

We adapt it to a surface we can reset. Each count is a fluorescent bead anchored by one molecular event, and the photothermal pulse is what releases the beads.

Digital PTR Assay Demonstration with DNA analyte

c09 - the cycle, running

Put the two together and this is what it looks like running.

The schematic is the duty cycle: measure, heat, recycle, measure. A hairpin on the surface and a hairpin on the bead are topologically locked until the target opens one of them by strand displacement, so a bead only anchors when the analyte is present.

The movie is the raw field at ten picomolar - single beads landing, then cleared. The trace underneath is the count over about twenty minutes: it climbs while analyte accumulates, and the pulse takes it back to the floor. The surface regenerates in about two minutes, and we have run it for at least thirteen cycles.

DNA is the easy case. It says the cycle itself works.

Digital PTR Assay Demonstration with DNA initiator

Tile Scanning

c10 - following a concentration up and down

Sensitivity is not the claim. Tracking is.

One field of view only holds so many binding sites, so we tile-scan: the stage walks a grid of fields and the counts add, which buys dynamic range without buying brightness. The limit of detection is seventy femtomolar, with about four decades of range - low femtomolar to high picomolar.

On the left, the initiator stepped up and down over about twelve hours. The sensor follows, and it comes back down. A sensor that only reports increases is an alarm. A monitor has to come back down.

On the right, the same experiment driven as a square wave: commanded input dashed, measured counts solid, with a photothermal reset between every step.

Detecting Nucleic Acid and Protein Analytes with Sandwich Assay

c11 - what the binder and the surface do to the reset

The reset is thermal, so anything that changes the binding energy changes the reset. We measured that rather than assuming it.

First the binder pair - antibody to antigen, DNA to DNA, nanobody to protein. All three reset and all three recover, so the format is not tied to one chemistry. The assay can follow whichever affinity reagent is best for a target.

Second, passivation. On a passivated surface the count returns cleanly to its floor after every pulse. Without it the residual creeps up, because what sticks non-specifically does not release at the same temperature as what binds specifically.

Detecting Nucleic Acid and Protein Analytes with Sandwich Assay

c12 - matrix and avidity

Two more variables, same experiment.

Matrix first: plasma, buffer and saliva behave the same through the cycle. The surface survives real fluids, with no bead accumulation on the gold.

Avidity is where it costs us. A two-site, high-avidity capture holds on through the pulse: that trace does not return to the same floor, and the gap widens over the run. That is the honest cost of the sandwich format, and it is why the small-molecule work in a moment moves to a one-site switch.

Protein Detection in Serum (Analyte: Thrombin)

Buffer

Serum (10%)

LOD 10 fM

LOD 31 fM

c14 - a protein, in serum

Thrombin, with an aptamer pair. The structure on the left is the aptamer on its target; under it, the sandwich that anchors the bead.

Top row is buffer: calibration, then tracking. Bottom row is the same two measurements in ten percent serum, the matrix that fouls most surfaces.

The curves barely move. Limit of detection is ten femtomolar in buffer and thirty-one in serum, and through stepped spike-ins the measured output sits on the input.

This is where the platform stops being a demonstration on DNA.

Detecting Small Molecule Analytes with Sandwich Assay

Intramolecular Strand Displacement (ISD)

Cortisol PTR sensor

ATP PTR sensor

LOD 107 fM

LOD 50 fM

c15 - small molecules need a one-site switch

A sandwich needs two binding sites, and a small molecule has one.

So the architecture changes: an intramolecular strand-displacement switch. The target binds the aptamer, a short complementary strand is displaced, the hairpin opens and the bead anchors. A one-site readout, and one whose thermodynamics can be tuned separately from its kinetics - the result Wilson and colleagues established.

Cortisol first: a hundred and seven femtomolar in buffer, eighty-six in artificial saliva, calibrated and then tracked over ten hours. Then ATP on the same switch chemistry: fifty femtomolar, and a twelve-hour run that follows the input across four decades, down and back up.

Same instrument, same reset, different recognition element. That is the argument for a platform rather than for any one assay.

Protein Binder Design with Generative AI Models

Target: osteocalcin (PDB 1Q8H)

b13k - the other route: design the binder

Every recognition element so far was bought or selected from a library. For the next target we designed it.

The target is osteocalcin, a bone-turnover marker: small, well studied, and it changes conformation with calcium, so there is a conserved epitope worth aiming at - the structure at lower left is the calcium-bound form. Above it is the assay this binder is for - the designed protein on the bead surface, paired with the aptamer, in the same digital PTR format you have been looking at.

On the right, what came out of the design pipeline: BindCraft generates candidate complexes, HADDOCK scores them, and the ten best went to the bench.

Binder Biochemical Characterization

binder 1K_D 0.11 nM

binder 2K_D 0.13 nM

binder 3K_D 0.29 nM

binder 4K_D 1.14 nM

binders 5-10no binding

OST aptamerK_D 0.21 nM

b13l - and they work

Ten designs on bio-layer interferometry, three or more replicates each.

Four bind, and bind well: a hundred and ten picomolar, a hundred and thirty, two hundred and ninety, and one point one nanomolar. The aptamer we had against the same target sits at two hundred and ten picomolar, so two of the designs beat it. Six did not pass.

Forty percent, from a few days of compute and one BLI run. Design proposes; measurement decides.

Synthetic Protein Binder for PTR Assay

Analyte: Osteocalcin

c17 - a binder we designed, in the assay

And into the assay.

The capture element here was not bought. It is one of those designed binders, clicked onto the bead surface and paired with the aptamer. Calibration first, then tracking: the sensor follows osteocalcin across four decades, down to about twenty femtomolar, and the lower trace catches the step.

Designed binder, thermal reset, digital count, one instrument.

Bacteria Culture Extracellular ATP Measurement

c16 - extracellular ATP, through a growth curve

Now the monitoring applications - three living systems, starting with the simplest.

An E. coli culture is plumbed through a selector valve and a syringe pump into the flow cell, so the sensor samples inline rather than from pulled aliquots - a few microlitres per assay, so the culture is not disturbed. Beside it, the same assay with and without the photothermal reset: without the reset the bead count drifts upward as beads accumulate; with it the count follows the input.

Underneath: extracellular ATP through fourteen hours of growth, with the optical density behind it. ATP peaks near the end of log phase and declines in stationary phase, which is the known physiology. The cross-check on the right puts PTR against a commercial luciferase assay on the same samples at three, four and five hours.

Two independent methods agreeing on a culture that was changing while we measured it. That is what a monitor does.

Long-Term Tissue-Chip Monitoring

Vitamin D3 addition

m10 - sixty hours on a live chip

The second living system is a tissue chip: human periodontal ligament stem cells in a hydrogel channel - the chip and the cells are on the top left.

Does the surface survive it? The trace at the bottom is sixty hours on a chip with cells in it, media fed continuously, the analyte spiked in so we know exactly what we asked for. Dashed is the command, black the measurement, grey the reading straight after the reset - the internal control, which stays on the floor the whole way. Four decades, up and down, for two and a half days.

And the measurement this was built for, top right: the same stem cells differentiating down the osteogenic path, with osteocalcin as the readout - the target our designed binder was made against. Calibration first, then the culture: vitamin D3 goes in on day three, and osteocalcin climbs from a few femtomolar to about a picomolar over the next five days, with the shape of that rise resolved because we sample twice a day.

Cortisol Monitoring

Lunch

Coffee

Dinner

Working

m14 - a day of cortisol

The last system is a person.

Hourly cortisol, from nine in the morning to one the following night. The blue boxes are the measurement; the grey ones beside them are the reading taken straight after the photothermal reset - the internal control, on the floor every time.

The trace has the shape it should have: high in the morning and falling through the afternoon. That is the diurnal rhythm, recovered by the sensor. And the day’s events sit on it - lunch, coffee, dinner, an evening of work.

Summary

• Photothermal recycling — a ~22 K pulse on a plasmonic Au film resets a tight binder in ~2 min, for ≥ 3 days, in plasma, serum, saliva and culture media
• Digital PTR assay — fM detection across ~4 decades for nucleic acids, proteins and small molecules, with DNA, aptamer, antibody, nanobody and designed-protein binders
• LOD: DNA 70 fM, thrombin 10 fM (31 in serum), cortisol 107 fM (86 in saliva), ATP 50 fM, osteocalcin 20 fM
• ~30 min per measurement, ~20 µL sample, > 60 h on chip
• Verification — inline eATP through an E. coli growth curve, osteocalcin from differentiating stem cells over multiple days

c26 - summary

To put it in one place.

Photothermal recycling turns a tight binder into a sensor that can be read again and again. A twenty-two-kelvin pulse on a gold film resets the surface in about two minutes, for at least a dozen cycles, in plasma, serum, saliva and culture media.

Combined with digital counting that reaches femtomolar - seventy for DNA, ten to thirty for thrombin, around a hundred for cortisol, fifty for ATP, twenty for osteocalcin - across four decades, with DNA, aptamer, antibody, nanobody and designed-protein binders. About thirty minutes per measurement, twenty microlitres of sample, and surfaces that last for days.

For a low-abundance analyte the reset is no longer the limit; the measurement speed is set by mass transport and association kinetics. That is the right problem for this community to have, and it is what we are working on next.

Acknowledgements

Thank you

Biodesign Lab

Yongchen Tai, Yunshen Li, Wenting Wang, Yida Wang

Mitchell Conover, Cindy Zhou, Jake Lee, Raymond Malanga, Aditi Kumari, Audrey Singletary

Collaborators

Wenjun Xie, Thomas Angelini, Qiye Zheng, Youngmin Park, Yonggang Ke, Jun-Chau Chien, Matthews Clayton, Xianrui Yang, Idalis Villanueva Alarcon

UF iGEM

2022-2026 teams

e06 - thank you

The people who did the work: Yongchen Tai, who led this, Yunshen Li, Yida Wang, Wenting Wang and Jacob Lee, with Qiye Zheng on the thermal measurements and our other collaborators.

Thank you - I am happy to take questions.