A PET isotope cyclotron becomes a proton radiobiology platform

What the work claims

This is a proof-of-concept facility paper: primary experimental results, but modest ones, whose purpose is to show that biological experiments at the Bern Medical Cyclotron are feasible and that the dosimetry chain is under control. The Bern Medical Cyclotron is an IBA Cyclone 18/18 HC, an 18 MeV compact machine built for microampere-range currents in routine commercial production of PET radiopharmaceuticals at Bern University Hospital. A beam transfer line of 6.5 meters ending in a separate research bunker lets physicists extract the proton beam away from the radioactive production targets, and a passive scattering arrangement can flatten the beam from over 1000 Gy per second at the high end down to below 0.05 Gy per second at the low end, spanning conventional and FLASH dose rates in one installation.¹ The beamline itself, including the beam profiler and the scattering components used for beam shaping, is documented in a companion methods paper from the same group.²

The biological claim is a pair of relative biological effectiveness values from clonogenic survival assays. Human keratinocytes (the HaCaT line) and mouse melanoma cells (B16-F10) were irradiated with 8.14 MeV protons at conventional dose rates and with 225 kV X-rays at doses from 0 to 8 Gy. Proton survival curves were steeper in both lines. Quantified at 0.1 survival, the relative biological effectiveness was 1.34 for the melanoma cells and 1.21 for the keratinocytes, both within the range the paper cites as typical for in vitro proton assays, where an average near 1.22 is usually quoted.¹

The facility claim is the bigger one, and the authors make it explicitly: to their knowledge the Bern machine is one of six facilities worldwide with documented setups for preclinical external-beam proton studies on a radiopharmaceutical cyclotron, and they state they are the first of the comparable retrofit facilities to report a calibrated irradiation setup covering both conventional and FLASH studies, with the highest extracted beam energy and the largest, most uniform extracted field in that peer group.¹

How it works

The physics is a study in what a production machine gives you for free and what it makes hard. The cyclotron delivers 18 MeV protons; extracted into air through the beamline the energy is 15.54 MeV, and after passing through the 1.5 mm face of a lucite cell flask it is 8.14 MeV at the cell layer, corresponding to a dose-weighted linear energy transfer of about 5.4 keV per micrometer inside the cells. Low energy simplifies collimation for future spatially fractionated minibeam work but means only thin, superficial samples can be irradiated: cell monolayers, or small superficial tumors in vivo. The 55 mm uniform field was achieved by scattering, at the cost of a 7.7 percent dose non-uniformity across the flask area, and the long exposures at conventional dose rates, averaging 1.9 Gy per minute, required an on-site incubator to hold the cultures at temperature.¹

The dosimetry chain is where the real intellectual work sits, because it is where a production instrument stops being a black box. Dose is monitored by a PTW type 786 ionization chamber calibrated against EBT-3 Gafchromic film, which sounds routine until you account for two proton-specific corrections. First, the films under-respond to low-energy protons in the Bragg peak region, a documented quenching effect worth a correction factor of 0.908. Second, the film's 125 micrometer polyimide lamination absorbs enough proton energy to raise the dose deposited in the film's active layer above what a 15 micrometer layer of cells actually receives, an overestimation the authors measured directly at 1.191 using partially covered films. Combined with beam non-uniformity and chamber calibration, the overall proton dose uncertainty comes to 12.8 percent, against a standard 5 percent assumed for the X-ray cabinet.¹

Biologically, survival was measured by the clonogenic assay, the oldest honest test in radiobiology: irradiate, incubate, count colonies of more than 50 cells after 9 days for the melanoma line and 14 days for the keratinocytes, and fit survival to the linear-quadratic model, whose alpha and beta parameters describe single-hit and accumulated damage respectively. All proton irradiations used flasks standing vertically with lids facing down, which during the multi-hour exposures let the culture medium drain into the lids and temporarily deprive the cells of it; the authors quantified that stress with a set of six control flasks held in the irradiation position for the longest exposure. The survival data come from two independent experiments with 12 flasks in total.¹

Where a skeptic should push

The most load-bearing number is the 12.8 percent proton dose uncertainty, and it should be read next to the headline results. The difference between the two relative biological effectiveness values, 1.34 versus 1.21, is smaller than the uncertainty in the proton dose scale itself. That does not invalidate the qualitative finding that protons out-killed X-rays in both lines, which is also the literature-consistent direction. It does mean the precision of the quoted effectiveness values is soft, and any downstream fractionation argument built on the fitted alpha-over-beta ratios, which moved from 3.2 to 1.09 Gy for the melanoma line, is hypothesis-grade at this sample size. The authors are careful here; a reader should be too.¹

The evidence base is two independent experiments, 12 flasks, two cell lines, one of which is an immortalized aneuploid human keratinocyte line derived decades ago from the histologically normal skin of a 62-year-old melanoma patient, the other a rodent metastatic melanoma line chosen partly for its radiosensitivity profile. In vitro, single-task, no replicates beyond two, no in vivo confirmation. The medium-deprivation confound is the detail a reviewer should linger on: proton exposures were so long that the culture setup itself stressed the cells, and while six control flasks bound that stress, the control measured survival after holding, not the interaction between medium deprivation and radiation damage. The authors note the in vitro to in vivo gap explicitly when discussing the melanoma fractionation hypothesis.¹

One more calibration is worth making because the facility context invites over-reading: nothing in this paper demonstrates FLASH or minibeam radiobiology. Every irradiation was at conventional dose rates. The machine's capability range, up to and beyond 1000 Gy per second, is asserted through beam physics, not exercised on cells here. That is the right order of operations for a new platform, but the headline applications are future work, and the paper says so.¹

When a production machine becomes a research platform

For platform access and the governance of experimentation on living tissue, the non-obvious implication is about where research capability comes from when no one will fund it directly. Nobody built the Bern facility as a radiobiology center; it exists because a commercial production asset, run jointly by a university, a hospital, and private investors, had a beamline, a spare bunker, and enough schedule slack to carry a second function. That retrofit pattern is spreading: the paper's own comparison table lists six facilities worldwide, including two others, in Spain and Portugal, adapting the same vendor's 18 MeV medical cyclotron in the same way. An installed base of hospital radioisotope machines is quietly becoming a dual-use research fleet, where the vendor capability that matters is not the catalog spec but how easily a production machine can be bent to research without breaking its revenue job.¹

That access model carries its own governance structure, and the paper documents it without naming it. Research time on a clinical production cyclotron is a scheduling negotiation with a machine whose primary duty is patient diagnostics; the facility's accessibility is real but permanently second-priority, and the six-facility table shows how geographically concentrated the capability is. Concentrated bespoke access has a well-known failure mode in every instrumentation field: results that depend on one machine's quirks, in one bunker, with one correction chain, do not transfer cleanly, and oversight has to be rebuilt locally at each site. The authors' insistence on publishing the full dosimetry uncertainty budget, quenching factors and lamination corrections included, is exactly the transparency habit that makes a dispersed retrofit fleet auditable instead of merely available, and it is the part of this paper other facilities should copy verbatim.¹

The connection to computing on living neural tissue is indirect but real, and it is metrological rather than biological. Platforms that run experiments on living substrate, organoids included, are converging on the same architecture this facility embodies: a commercial instrument, a custom interfacing layer, and a chain of corrections that translates what the instrument measures into what the biology experienced. The Bern paper shows that the correction chain is the governance surface. A 12.8 percent dose uncertainty does not just soften a survival curve; it sets the floor on what any safety claim built downstream of that instrument can honestly say. The medium-deprivation episode is the same lesson in welfare terms: an experimental convenience introduced by the hardware geometry became an uncontrolled stressor on the living substrate, caught only because someone thought to run six control flasks. On a closed-loop platform running thousands of automated exposures, there will be no one to notice unless the control logic is designed to notice. The opportunity is that retrofit instrumentation, published with this level of metrological candor, gives small laboratories access to capabilities that used to require national facilities. The threat is that the corrections stay in the methods section, while the conclusions and the procurement decisions inherit none of the doubt.¹

The bottom line

Established by this paper: a commercial PET cyclotron can be converted into a calibrated preclinical proton radiobiology platform, and its first in vitro results are directionally consistent with decades of radiobiology, with relative biological effectiveness values of 1.34 and 1.21 in two cell lines. Not yet established: anything about FLASH, minibeam, in vivo response, or fractionation strategy, all of which remain designs and hypotheses. The durable lesson for the living-tissue platform field is that scarce experimentation capability is increasingly arriving as a retrofit with a published uncertainty budget, and the quality of that budget, not the capability itself, is what determines how much trust the downstream science deserves. What would confirm the facility's promise is a FLASH-dose-rate biological result with the same rigor; what would break the model is evidence that retrofit access cannot reproduce results across the six-site fleet, which would turn each facility into its own epistemic island.