PhotoFiTT: A Quantitative Framework for
Assessing Phototoxicity in Live-Cell
Microscopy Experiments
Mario Del Rosario1,*, Estibaliz Gómez-de-Mariscal1,*, Leonor Morgado1,2, Raquel Portela3, Guillaume Jacquemet4,5,6,7, Pedro
M. Pereira3,
, and Ricardo Henriques1,8,
*Equally contributed authors
1Optical cell biology group, Instituto Gulbenkian de Ciência, Oeiras, Portugal
2Abbelight, Cachan, France
3Intracellular Microbial Infection Biology Laboratory, Instituto de Tecnologia Química e Biológica António Xavier, Universidade Nova de Lisboa, Portugal
4Turku Bioimaging, University of Turku and Åbo Akademi University, Turku, Finland
5Faculty of Science and Engineering, Cell Biology, Åbo Akademi University, Turku, Finland
6InFLAMES Research Flagship Center, Åbo Akademi University, Turku, Finland
7Turku Bioscience Centre, University of Turku and Åbo Akademi University, 20520, Turku, Finland
8MRC Laboratory for Molecular Cell Biology, University College London, London, United Kingdom
Phototoxicity in live-cell fluorescence microscopy can compro-
mise experimental outcomes, yet quantitative methods to assessits impact remain limited. Here we present PhotoFiTT (Photo-toxicity Fitness Time Trial), an integrated framework combin-
ing a standardised experimental protocol with advanced image
analysis to quantify light-induced cellular stress in label-free set-tings. PhotoFiTT leverages machine learning and cell cycle dy-namics to analyse mitotic timing, cell size changes, and overallcellular activity in response to controlled light exposure. Usingadherent mammalian cells, we demonstrate PhotoFiTT’s abilityto detect wavelength- and dose-dependent effects, showcasing
that near-UV light induces significant mitotic delays at doses
as low as 0.6J/cm
2, while longer wavelengths require higher
doses for comparable effects. PhotoFiTT enables researchers to
establish quantitative benchmarks for acceptable levels of pho-
todamage, facilitating the optimisation of imaging protocols thatbalance image quality with sample health.
phototoxicity | microscopy | machine learning | cell division | cellular activity
Correspondence: (P. M. Pereira) pmatos@itqb.unl.pt, (R. Henriques) rjhen-
riques@igc.gulbenkian.pt
Main
Live-cell fluorescence microscopy has revolutionised our
ability to study dynamic biological processes in near-native
conditions (1). However, the intense illumination required
can induce phototoxicity, disrupting cellular functions (2–
5). These effects, often subtle and cumulative, can leadto significant changes in cell behaviour, organelle integrity,and developmental processes (6–9). Yet, establishing opti-
mal imaging protocols that balance high-quality data acqui-
sition with minimal biological interference remains a com-
plex challenge (10). Traditional methods for assessing pho-totoxicity include viability assays and morphological ob-servations. Historically, photobleaching has been used as
a proxy reporter of phototoxicity, but with the advent of
highly photostable fluorescent proteins like mStayGold (11),
this approach has become less reliable. Importantly, tol-erance to photodamage varies across specimens and is in-fluenced by damage severity, complicating the development
Fig. 1. Phototoxicity fitness time trial (PhotoFiTT) assay. PhotoFiTT inte-
grates a biological imaging assay with an image analysis workflow, using cellu-lar processes and mitotic cycles as natural timers to track consistent patterns of
cell behaviour. It detects deviations in these patterns caused by light exposure,
enabling the quantification of phototoxic effects in fluorescence microscopy. The
workflow analyses three cellular features: a) Mitosis: Identifies mitotic rounding
(yellow squares), typically 30-50 minutes post-G2 exit. Phototoxicity alters round-
ing time distributions (orange: normal, blue: photodamaged). b) Size: Tracks cell
diameters (black arrowheads) through division. Mother cells (purple) transition to
daughter cells (orange) approximately
60minutes post G2 exit. Phototoxicity delays
this transition (light orange). c) Activity: Quantifies sub-cellular changes between
frames (red outlines). Normal cells show increased post-division activity (orange,
around 60 minutes), while photodamaged cells (blue) exhibit reduced activity.
of replicable imaging protocols (12, 13). Previous studies
have offered valuable quantitative assessments of phototoxic-
ity (3, 10,14), but a critical need remains for standardised and
generalisable methods able to quantitatively link cell damage
to high-intensity light exposure across different fluorescence
Del Rosario & Gómez-de-Mariscal et al. | bioR ‰iv | July 16, 2024 | 1–14. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
microscopy techniques. Addressing this gap, we introduce
PhotoFiTT (Phototoxicity Fitness Time Trial), a quantitativeimaging-based framework designed to assess phototoxicity
effects on cellular behaviour in live-cell microscopy exper-
iments (Figure 1). PhotoFiTT provides a rigorous, label-
free, and quantitative approach to evaluate phototoxicity, en-hancing the reliability and reproducibility of live-cell imag-ing studies. It is designed as an integrated framework com-
prising a standardised experimental protocol and an advanced
image analysis pipeline. PhotoFiTT leverages the predictablenature of cell division as a "biological clock" to quantify
phototoxicity-induced perturbations (Figure 1). The frame-
work uses low-illumination brightfield microscopy to moni-
tor cell populations that have been exposed to controlled light
damage events. It employs deep learning algorithms, includ-ing virtual staining and cell segmentation, to aid in tracking
abnormal light-induced cellular behaviour. PhotoFiTT em-
ploys three key measurements: 1) Mitosis monitoring: as-
sess delays in cell division by mapping mitotic cell roundingin time (Figure 1a); 2) Cell size dynamics: track changes
in cell size to discriminate between cell division and cell cy-cle arrest (Figure 1b); 3) Cellular activity: quantify sub-
cellular changes over time as a measure of overall cellular
health (Figure 1c). By analysing these factors, PhotoFiTT
enables researchers to extract numerical constraints for opti-
mising live-cell-compatible illumination conditions.
To implement PhotoFiTT (Figure 2a), researchers can fol-
low these key steps: 1) Synchronise cells using the CDK-1
inhibitor RO-3306, which arrests cells in the G2/M interface,
or study unsynchronised cells that can be tracked from mi-
totic rounding until division; 2) Expose cells to a light irradi-
ation event replicating the illumination pattern of the imaging
experiment to be analysed for phototoxity; 3) Identify cellsundergoing mitotic rounding and division (Figure 2b) using
a Stardist (15) deep learning model pursposely-trained for
brightfield round cell detection; 4) Analyse three key param-
eters: mitotic timing, cell diameter dynamics, and cellular
activity. Video S1-2 provide in-detail tutorials to reproduce
these procedures. We conducted a comprehensive analysis of
phototoxicity using Chinese Hamster Ovary (CHO) cells as a
model system. CHO cells have the benefit of having a short
doubling time (14-17 hours), allowing for efficient experi-
mental timelines, while being a highly popular cell line for
microscopy and cell research. Both synchronised and unsyn-chronised cell populations were exposed to varying doses ofnear-UV (385 nm), blue (475 nm), and red (630 nm) light to
assess wavelength-dependent effects. In synchronised cells,
near-UV irradiation induced significant dose-dependent de-
lays in mitotic timing, with effects detectable at doses aslow as 0.6J/cm
2(Figure 2c-d, Video S3). Notably, at
60J/cm2, the characteristic peak in mitotic rounding be-
came almost imperceptible, indicating widespread cell cycle
arrest. To validate that these effects were primarily due to
light exposure rather than the potential light sensitivity of the
CDK1 inhibitor used for synchronisation, we conducted par-allel experiments with unsynchronised cells. In unsynchro-nised cell populations exposed to near-UV light, we observeda dose-dependent decrease in dividing cells and a simultane-ous increase in arrested cells (Figure S1a). Consistent with
our findings in synchronised populations, the time required
for cells to divide after entering mitotic rounding increased
with higher light doses (Figure S1b). These results confirm
that the observed effects were primarily attributable to lightexposure rather than light sensitivity increase by the synchro-nisation drug. Notably, synchronised populations exhibitedhigher sensitivity to near-UV-induced damage compared to
their unsynchronised counterparts. This increased suscepti-
bility was evidenced by a higher proportion of apoptotic cells,
as identified by a SYTOX-based cell viability assay (FigureS1c). This observation can be attributed to the fact that syn-chronised cells were uniformly arrested at the G2/M check-
point, a critical stage immediately following DNA replica-
tion and preceding mitosis. Near-UV radiation is known to
induce DNA damage (12), and cells in G2 phase are par-ticularly vulnerable due to the presence of fully replicatedchromosomes and the imminent onset of mitosis. The accu-mulation of DNA damage at this stage can trigger cell cycle
arrest or apoptosis more readily than in unsynchronised pop-
ulations, where cells are distributed across various cell cy-cle phases (16). While near-UV irradiation demonstrated themost pronounced effects, longer wavelengths also inducedcellular stress at higher doses. Cells exposed to 475nmand
630nmlight began showing comparable delays to those ob-
served with near-UV light (i.e., mitotic rounding time pointsabove 60min) only at the highest tested dose of 60J/cm
2
(Figure 2d). This finding highlights a clear wavelength de-
pendence in phototoxicity, with longer wavelengths requir-ing substantially higher doses to induce comparable photo-
damage. PhotoFiTT’s analysis of cell size dynamics revealed
that near-UV exposure delayed the appearance of daughter
cells in a dose-dependent manner (Figure 2e). Under con-
trol conditions, mother cells (≥ 20µm diameter) divide into
daughter cells (≥ 15µm diameter) within 50≠75minutes
post-synchronisation. Near-UV light exposure induced dose-dependent delays in this process, with high doses (60J/cm
2)
delaying daughter cell appearance by up to 120minutes after
synchronisation. Such delay in division and the persistence
of mother cells upon time indicates the increase in the num-
ber of cells arrested due to photodamage and explains the lostpeak in the distribution of mitotic rounding with high doses
(60J/cm
2) inFigure 2c. Classifying mother and daugh-
ter cells according to their size lets PhotoFiTT evaluate cell
division delays (Figure 2f). Upon challenged with longer
wavelengths (475, and 630nm) and in agreement with our
previous observations (Figure 2d), cell division delays are
also wavelength-dependant, requiring higher light doses to
induce noticeable effects (Figure 2f). Beyond acute effects
on cell division, PhotoFiTT can also quantify post-mitotic
cellular activity, providing insight into long-term impact on
cellular behaviour. We observe that higher light doses andshorter wavelengths led to decreased cumulative cellular ac-tivity over a 7-hour period (Figure 2g). While each of these
metrics has its own benefits, tracking the 1-to-2 transition of a
mother cell to two resolvable daughter cells provides a highly
2 | bioR ‰iv Del Rosario & Gómez-de-Mariscal et al. | PhotoFiTT: Quantifying Live-Cell Phototoxicity. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
Fig. 2. Quantitative phototoxicity assessment with PhotoFiTT. a) PhotoFiTT workflow combines biochemical experiments, microscopy image acquisition and image
analysis. Cell cycle synchronisation is induced by a CDK1 inhibitor (RO-3306) arresting cells at the G2/M interface. Photodamage irradiation represents the desired
illumination patterns to be studied. Synchronised cells are washed from the drug used prior to acquisition. Time-lapse acquisition starts, using transmitted light, lasting 6
hours. A small subset of the videos is manually annotated to train a cell mitotic rounding identification model. The main experimental time-lapse videos are processed to
identify the proportional mitotic rounding with the trained models and calculate the cell activity across time. These measurements are used for the final data analysis. Detailed
workflow description Note S1 and reproducibility guidelines Note S2 are given in the Supplementary Information and Methods section. b) Time-lapse brightfield imaging
of an adherent mammalian cell (Chinese hamster ovary - CHO) undergoing mitosis. The mother cell’s diameter approximately doubles that of the resulting daughter cells
post-division. Scale bar 20µm. c) Temporal distribution of mitotic cell rounding in synchronised populations exposed to varying doses of 385nm (near-UV) light ( 0.6,6,
and60J/cm2). Exposed populations exhibit a dose-dependent delay in mitotic rounding, manifested as rightward shifts in the distribution peaks. d) Quantification of mitotic
rounding delays across different light wavelengths and doses. The control population peaks at t50minutes, representing the average mitotic rounding time (indicated with
a line in b). Exposure to 385nm light induces a dose-dependent delay, while 475nm and630nm wavelengths only cause significant delays at high doses ( 60J/cm2).
e) Cell volume nearly halves as mother cells divide into two daughter cells. Low-dose exposure ( 0.6J/cm2) delays this transition, while high doses result in heterogeneous
populations at later time points ( 120min ), indicating asynchronous divisions and potential cell cycle arrest. f) Temporal analysis of mother and daughter cell populations
across different wavelengths and doses. All wavelengths induce delays in daughter cell appearance, with 385nm light and high doses of 475nm and630nm light causing
the most pronounced effects. g) Cumulative cell activity over 7hhours post-exposure normalised for each replica. All light exposures reduce overall cell activity, with 385nm
and475nm showing more dose-dependent effects than 630nm light. These results collectively demonstrate the wavelength- and dose-dependent impacts of light exposure
on cell division dynamics and overall cellular activity.
Del Rosario & Gómez-de-Mariscal et al. | PhotoFiTT: Quantifying Live-Cell Phototoxicity bioR‰iv | 3. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
efficient and straightforward approach to assessing photo-
damage effects. This transition can be easily monitored byclassifying cell rounding events pre- and post-division within
a short time window (15 ≠30min). In control, synchronised
populations, approximately 50% of cells transition from re-
solvable mother to daughter cells within the first 50minutes
post-synchronisation. At a light dose of 6J/cm
2, the pro-
portion of cells completing the mother-to-daughter transition
within this time-frame is reduced to around 40% for630nm
illumination, 30% for475nm, and only 10% for405nm
(Figure 2e-f). Taken together, our results demonstrate the
nuanced and continuous nature of phototoxicity. They reveala spectrum of cellular responses that vary with both light doseand wavelength, underscoring the importance of considering
them in experimental design.
Our findings reveal that phototoxicity not only induces
delays in cell division but also compromises subsequent
cellular functions ( Figure S2c-g), corroborating and extend-
ing previous observations. The damage inflicted by shorter
wavelengths is more pronounced, likely due to their knowndirect interaction with DNA (17). Longer wavelengths, whilegenerally less harmful, have been described to exert signif-
icant effects on cellular physiology, including alterations in
mitochondrial membrane potential, cytoskeletal dynamics,
and reactive oxygen species levels (3, 18). The observed
physiological responses likely stem from a complex inter-play of multiple cellular perturbations, rather than isolated
phototoxic effects. This multifaceted nature of phototoxicity
underscores the importance of comprehensive assessment
approaches in live-cell imaging experiments. A key strength
of PhotoFiTT is its ability to reveal the non-binary nature ofphototoxicity, demonstrating a continuum of effects that vary
with light dose and wavelength in a label free manner. This
offers advantages over simple viability assays (Figure S1c)
and enables fine assessment of specific illumination condi-
tions with minimal interference from the method itself. By
iteratively applying PhotoFiTT with different illumination
parameters, researchers can identify optimal settings that
balance image quality with minimal phototoxicity. Thisapproach allows for fine-tuning of imaging protocols based
on quantitative data rather than qualitative assessments.
Researchers can systematically vary labels, light doses,wavelengths, and exposure times, then use PhotoFiTT to
measure the resulting effects on cell division timing, cell size
dynamics, and overall cellular activity. Using PhotoFiTTresults to establish quantitative benchmarks for acceptable
levels of photodamage, researchers can make informed
decisions about imaging parameters and more accurately
interpret their results. This is particularly important given
the growing recognition of the "reproducibility crisis"in biomedical research (19). Additionally, PhotoFiTTmeasurements enable comprehensive tracking of cells’ phys-
iology deviations, making it suitable as an early-warning
system for cellular stress. It is important to acknowledge
limitations of PhotoFiTT. While highly sensitive to majorcellular perturbations, it may not detect subtle light-inducedevents. Its primary strength lies in identifying conditionsthat significantly impact cell division and overall cellularhealth without requiring fluorescent labels. PhotoFiTT isdesigned for adherent, dividing cell lines, which may limit
its applicability to other experimental systems. Our results
underscore the importance of considering the potentially
damaging effects of activation light in super-resolution tech-niques such as Single Molecule Localisation Microscopy(SMLM) and Single Particle tracking (SPT), which often useillumination in the range of kW/cm
2for several minutes.
For imaging domains where illuminations are well abovethe60J/cm
2threshold studied here, researchers should
carefully consider and mitigate photodamage. Future workcould explore the development of accelerated protocols
or the integration of PhotoFiTT principles into real-time
analysis pipelines to apply corrective measures to avoid cell
death. The principles underlying PhotoFiTT could also be
extended to other imaging modalities or adapted for specificbiological questions, potentially allowing for real-time
adjustment of imaging parameters to minimise photodamage
during long-term experiments.
DATA AVAILABILITY
The data obtained in this study, as well as annotated images are avail-
able through the BioArchive at https://www.ebi.ac.uk/biostudies/
bioimages/studies/S-BIAD1269 and example images to test the code are
available through https://zenodo.org/records/12733476 both under CC
BY 4.0 license.
CODE AVAILABILITY
The PhotFiTT computational framework is available at https://github.com/
HenriquesLab/PhotoFiTT . We used ZeroCostDL4Mic to train and assess
the StarDist and Pix2Pix models: https://github.com/HenriquesLab/
ZeroCostDL4Mic . All source code is under an MIT License.
AUTHOR CONTRIBUTIONS
R.H. and P .M.P . conceived the project. M.DR., E.G.M. P .M.P . and R.H. designed the
experimental and analytical pipelines. M.DR. and L.M. performed the live imagingexperiments. M.DR., E.G.M. and R.P . acquired the brightfield-fluorescent paireddata. M.DR., E.G.M. and L.M. annotated and curated the data. E.G.M. developedthe analytical pipeline. M.DR. wrote the Fiji macros. E.G.M and M.DR. analysedthe data with inputs from G.J., P .M.P . and R.H.. M.DR. and E.G.M. prepared andtested the GitHub repository and wrote the paper with input from G.J., P .M.P . andR.H. All authors reviewed and refined the manuscript.
ACKNOWLEDGEMENTS
We thank Kalina Tosheva, Buzz Baum and Caren Norden for discussions regard-
ing experimental workflows. M.DR., E.G.M. and R.H. acknowledge the support of
the Gulbenkian Foundation (Fundação Calouste Gulbenkian), the European Re-
search Council (ERC) under the European Union’s Horizon 2020 research and in-
novation programme (grant agreement No. 101001332) (to R.H.) and funding from
the European Union through the Horizon Europe program (AI4LIFE project withgrant agreement 101057970-AI4LIFE and RT -SuperES project with grant agree-ment 101099654-RTSuperES to R.H.). Funded by the European Union. Viewsand opinions expressed are, however, those of the authors only and do not nec-essarily reflect those of the European Union. Neither the European Union nor thegranting authority can be held responsible for them. This work was also supportedby a European Molecular Biology Organization (EMBO) installation grant (EMBO-
2020-IG-4734 to R.H.), an EMBO postdoctoral fellowship (EMBO ALTF 174-2022
to E.G.M.), a Chan Zuckerberg Initiative Visual Proteomics Grant (vpi-0000000044
with https://doi.org/10.37921/743590vtudfp to R.H.) and a Chan Zuckerberg Initia-tive Essential Open Source Software for Science (EOSS6-0000000260). A joint-collaboration between Abbelight and the Instituto Gulbenkian de Ciência kindlysupports L.M. R.H. and P .M.P acknowledge funding by Fundação para a Ciên-cia e Tecnologia (FCT, Portugal) through national funds to the Associate Labo-ratory LS4FUTURE (LA/P/0087/2020, DOI 10.54499/LA/P/0087/2020). P .M.P . ac-knowledges support from Fundação para a Ciência e Tecnologia (Portugal) project
grant (PTDC/BIA-MIC/2422/2020), the R&D unit Mostmicro (UIDB/04612/2020,
UIDP/04612/2020), La Caixa Junior Leader Fellowship (LCF/BQ/PI20/11760012)financed by ”la Caixa” Foundation (ID 100010434) and by European Union’s Hori-zon 2020 research and innovation programme under the Marie Skłodowska-Curiegrant agreement No 847648, and a Maratona da Saúde award. This study was also
supported by the Research Council of Finland (338537 to G.J.), the Sigrid Juselius
Foundation (to G.J.), the Cancer Society of Finland (Syöpäjärjestöt; to G.J.), andthe Solutions for Health strategic funding to Åbo Akademi University (to G.J.). Thisresearch was supported by the InFLAMES Flagship Programme of the Academy of
4 | bioR
‰iv Del Rosario & Gómez-de-Mariscal et al. | PhotoFiTT: Quantifying Live-Cell Phototoxicity. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
A Excitation light irradiation
Finland (decision numbers: 337530, 337531, 357910, and 357911).
EXTENDED AUTHOR INFORMATION
•M. Del Rosario: 0000-0002-0430-1463; ￿_mariod
•E. Gómez-de-Mariscal: 0000-0003-2082-3277; ￿gomez_mariscal
•L. Morgado: 0000-0002-3760-5180; ￿ALeonorMorgado
•R. Portela: 0000-0002-5559-9554
•G. Jacquemet: 0000-0002-9286-920X; ￿guijacquemet
•P . M. Pereira: 0000-0002-1426-9540; ￿P_Matos_Pereira
•R. Henriques: 0000-0002-2043-5234; ￿HenriquesLab
Bibliography
1. Jeff W Lichtman and José-Angel Conchello. Fluorescence microscopy. Nature Methods,2
(12):910–919, December 2005. ISSN 1548-7091, 1548-7105. doi: 10.1038/nmeth817 .
2. Jaroslav Icha, Michael Weber, Jennifer C. Waters, and Caren Norden. Phototoxicity in live
fluorescence microscopy, and how to avoid it. BioEssays: News and Reviews in Molecular,
Cellular and Developmental Biology, 39(8), August 2017. ISSN 1521-1878. doi: 10.1002/
bies.201700003 .
3. Sina Wäldchen, Julian Lehmann, Teresa Klein, Sebastian van de Linde, and Markus Sauer.
Light-induced cell damage in live-cell super-resolution microscopy. Scientific Reports, 5(1):
15348, October 2015. ISSN 2045-2322. doi: 10.1038/srep15348 . Number: 1 Publisher:
Nature Publishing Group.
4. Michael A. Reiche, Jesse S. Aaron, Ulrike Boehm, Michael C. DeSantis, Chad M. Hobson,
Satya Khuon, Rachel M. Lee, and Teng-Leong Chew. When light meets biology – how the
specimen affects quantitative microscopy. Journal of Cell Science, 135(6):jcs259656, March
2022. ISSN 0021-9533. doi: 10.1242/jcs.259656 .
5. P . Philippe Laissue, Rana A. Alghamdi, Pavel Tomancak, Emmanuel G. Reynaud, and Hari
Shroff. Assessing phototoxicity in live fluorescence imaging. Nature Methods, 14(7):657–
661, July 2017. ISSN 1548-7105. doi: 10.1038/nmeth.4344 . Number: 7 Publisher: Nature
Publishing Group.
6. Abhilasha Ojha and Navin Kumar Ojha. Excitation light-induced phototoxicity during fluo-
rescence imaging. Journal of Biosciences, 46:78, 2021. ISSN 0973-7138.
7. Xinyi Zhang, Gabriel Dorlhiac, Markita P . Landry, and Aaron Streets. Phototoxic ef-
fects of nonlinear optical microscopy on cell cycle, oxidative states, and gene expres-
sion. Scientific Reports, 12(1):18796, November 2022. ISSN 2045-2322. doi: 10.1038/
s41598-022-23054-7 . Number: 1 Publisher: Nature Publishing Group.
8. Andreas P . Cuny, Fabian P . Schlottmann, Jennifer C. Ewald, Serge Pelet, and Kurt M.
Schmoller. Live cell microscopy: From image to insight. Biophysics Reviews, 3(2):021302,
June 2022. doi: 10.1063/5.0082799 . Publisher: American Institute of Physics.
9. Petr Strnad, Stefan Gunther, Judith Reichmann, Uros Krzic, Balint Balazs, Gustavo
de Medeiros, Nils Norlin, Takashi Hiiragi, Lars Hufnagel, and Jan Ellenberg. Inverted light-sheet microscope for imaging mouse pre-implantation development. Nature Methods, 13
(2):139–142, February 2016. ISSN 1548-7091, 1548-7105. doi:
10.1038/nmeth.3690 .
10. Estibaliz Gómez-de Mariscal, Mario Del Rosario, Joanna W. Pylvänäinen, Guillaume
Jacquemet, and Ricardo Henriques. Harnessing artificial intelligence to reduce phototox-icity in live imaging. Journal of Cell Science, 137(3):jcs261545, February 2024. ISSN
0021-9533, 1477-9137. doi:
10.1242/jcs.261545 .
11. Ryoko Ando, Satoshi Shimozono, Hideo Ago, Masatoshi Takagi, Mayu Sugiyama, Hiroshi
Kurokawa, Masahiko Hirano, Yusuke Niino, Go Ueno, Fumiyoshi Ishidate, et al. Staygoldvariants for molecular fusion and membrane-targeting applications. Nature Methods, 21(4):
648–656, 2024.
12. Maria Zhivagui, Areebah Hoda, Noelia Valenzuela, Yi-Yu Y eh, Jason Dai, Yudou He, Shu-
vro P . Nandi, Burcak Otlu, Bennett Van Houten, and Ludmil B. Alexandrov. DNA damage andsomatic mutations in mammalian cells after irradiation with a nail polish dryer. Nature Com-
munications, 14(1):276, January 2023. ISSN 2041-1723. doi:
10.1038/s41467-023-35876-8 .
13. Frederic Strobl and Ernst H. K. Stelzer. Non-invasive long-term fluorescence live imaging
ofTribolium castaneum embryos. Development, 141(11):2331–2338, June 2014. ISSN
1477-9129, 0950-1991. doi: 10.1242/dev.108795 .
14. Jean-Yves Tinevez, Joe Dragavon, Lamya Baba-Aissa, Pascal Roux, Emmanuelle Perret,
Astrid Canivet, Vincent Galy, and Spencer Shorte. Chapter fifteen - A Quantitative Method
for Measuring Phototoxicity of a Live Cell Imaging Microscope. In P . Michael Conn, edi-
tor,Methods in Enzymology, volume 506 of Imaging and Spectroscopic Analysis of Living
Cells, pages 291–309. Academic Press, January 2012. doi: 10.1016/B978-0-12-391856-7.
00039-1 .
15. Uwe Schmidt, Martin Weigert, Coleman Broaddus, and Gene Myers. Cell Detection with
Star-convex Polygons. 2018. doi: 10.48550/ARXIV .1806.03535 . Publisher: arXiv Version
Number: 2.
16. Véronique Gire and Vjekoslav Duli ´c. Senescence from G2 arrest, revisited. Cell Cycle , 14
(3):297–304, February 2015. ISSN 1538-4101, 1551-4005. doi: 10.1080/15384101.2014.
1000134 .
17. Kalina L. Tosheva, Yue Yuan, Pedro Matos Pereira, Siân Culley, and Ricardo Henriques.
Between life and death: strategies to reduce phototoxicity in super-resolution microscopy.Journal of Physics D: Applied Physics, 53(16):163001, February 2020. ISSN 0022-3727.doi:
10.1088/1361-6463/ab6b95 . Publisher: IOP Publishing.
18. Jaroslav Icha, Christiane Kunath, Mauricio Rocha-Martins, and Caren Norden. Indepen-
dent modes of ganglion cell translocation ensure correct lamination of the zebrafish retina.Journal of Cell Biology , 215(2):259–275, October 2016. ISSN 0021-9525, 1540-8140. doi:
10.1083/jcb.201604095 .
19. Monya Baker. 1,500 scientists lift the lid on reproducibility. Nature, 533(7604):452–454,
May 2016. ISSN 0028-0836, 1476-4687. doi: 10.1038/533452a .
20. Phillip Isola, Jun-Y an Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation
with conditional adversarial networks. In Proceedings of the IEEE conference on computer
vision and pattern recognition, pages 1125–1134, 2017.21. Juan C Caicedo, Allen Goodman, Kyle W Karhohs, Beth A Cimini, Jeanelle Ackerman,
Marzieh Haghighi, CherKeng Heng, Tim Becker, Minh Doan, Claire McQuin, et al. Nucleussegmentation across imaging experiments: the 2018 data science bowl. Nature methods,
16(12):1247–1253, 2019.
22. Lucas von Chamier, Romain F . Laine, Johanna Jukkala, Christoph Spahn, Daniel Krentzel,
Elias Nehme, Martina Lerche, Sara Hernández-Pérez, Pieta K. Mattila, Eleni Karinou, Séa-
mus Holden, Ahmet Can Solak, Alexander Krull, Tim-Oliver Buchholz, Martin L. Jones,
Loïc A. Royer, Christophe Leterrier, Y oav Shechtman, Florian Jug, Mike Heilemann, Guil-laume Jacquemet, and Ricardo Henriques. Democratising deep learning for microscopywith ZeroCostDL4Mic. Nature Communications , 12(1):2276, December 2021. ISSN 2041-
1723. doi:
10.1038/s41467-021-22518-0 .
23. Lyubomir T Vassilev. Cell Cycle Synchronization at the G 2/M Phase Border by Reversible
Inhibition of CDK1. Cell Cycle , 5(22):2555–2556, November 2006. ISSN 1538-4101, 1551-
4005. doi: 10.4161/cc.5.22.3463 .
24. Iván Hidalgo-Cenalmor, Joanna W Pylvänäinen, Mariana G. Ferreira, Craig T Russell, Alon
Saguy, Ignacio Arganda-Carreras, Y oav Shechtman, Guillaume Jacquemet, Ricardo Hen-riques, et al. Dl4miceverywhere: deep learning for microscopy made flexible, shareable andreproducible. Nature Methods, pages 1–3, 2024.
Methods
Cell lines. CHO (ATCC CCL-61) cells were cultured in
DMEM (Gibco) supplemented with either 42µM gentam-
icin (Gibco) or 1%penicillin/streptomycin (Gibco) and 10%
foetal bovine serum (FBS; Gibco). All cells were grown at
37ºC in a 5%CO 2humidified incubator.
Cell cultures, synchronisation. Cells were seeded on an
8well-chambered cover glass (Cellvis) with a total of 6x104
cells per well. If cell synchronisation was required, cells wereincubated with 10nM RO-3306 (Sigma-Aldrich) to inhibit
CDK1 activity for 16≠18hours. Unsynchronised cells were
seeded using the same protocol and allowed to grow for the
same duration.
Cell fixation and nuclei staining to generate fluores-
cent and brightfield paired images. Cells grown on a 8
well-chambered cover glass (Cellvis) until 60% confluency
were fixed with a 4%paraformaldehyde solution (Electron
Microscopy Sciences) for 10 minutes and washed three times
with PBS. Subsequently, the samples were incubated with a
0.1µg/ml Hoechst 33342 solution for 10 minutes in the dark.
A. Excitation light irradiation. Cells were then exposed to
excitation light at the Axio Observer 7 (Zeiss) microscopeequipped with a Colibri 7 LED illumination (Zeiss). 10fields
of view (FOV) were randomly selected per exposure condi-tion. The exposure time was set for different fields of view inthe same well according to the experimental conditions (100ms corresponding to 0.6J/cm
2,1s to6J/cm2and10s to
60J/cm2). The power used for each wavelength line was
the following: 84.9 ±2.8mW for the 385nmline, 85.2 ±
0.8for the 475nm, and 70.6 ±0.4for630nm(mean ±
standard deviation). Exposed areas were carefully selectedso as not to overlap. Immediately after light exposure and be-fore imaging, cells were washed two times with PBS and RO-3306 containing media was replaced by phenol-red free Fluo-robrite DMEM (Gibco) supplemented with 2 mM GlutaMAX
(Gibco), 42µM gentamicin (Gibco) and 10% FBS (Gibco).
The media exchange was done with a custom-made 4-syringe
holder designed in TinkerCAD (https://www.tinkercad.com/)and 3D printed using a Prusa MK4 3D printer (.stl file in sup-plementary material).
Del Rosario & Gómez-de-Mariscal et al. | PhotoFiTT: Quantifying Live-Cell Phototoxicity bioR‰iv | 5. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
B. Live-cell imaging and generation of fluorescent
and brightfield paired images for training. Live-cell
imaging was performed on the Axio Observer 7 (Zeiss) mi-
croscope equipped with a Colibri 7 LED illumination (Zeiss),
a20x/0.8 Plan Apochromat objective (Zeiss) and a Prime
95B sCMOS camera (Teledyne Photometrics). Images were
acquired every 4minutes for 8hours and have a pixel size
of0.55µm ◊0.55µm. The temperature and CO 2levels were
maintained at 37ºC and 5%respectively throughout the ex-
citation light exposure and imaging periods. For fixed sam-
ples, cells were imaged using the same objective in a multi-
channel acquisition mode. Brightfield images were captured
alongside fluorescent images using the 385nm filter to obtain
paired brightfield and Hoechst-stained images.
C. Live-cell viability. Cell viability was performed by
adding 0.1µM SYTOX Orange (Thermo) to previously
seeded cells and taken to the microscope for live-cell imag-
ing for 12hours. Live-cell imaging was performed on the
Nikon Eclipse Ti2 microscope equipped with a CoolLED
pe800 illumination (Nikon), a 20x/0.8 Plan Apochromat ob-jective (Nikon) and an Orca-fusionBT camera (Hamamatsu).10fields of view (FOV) were randomly selected per exposure
condition and the images were acquired every 4minutes us-
ing the brightfield and TRITC channel in a multichannel ac-quisition. The temperature and CO
2levels were maintained
at37ºC and 5%.
D. General image processing workflow. As depicted in
Figure 2, the image analysis to extract the measurements of
interest is composed of: (1) detecting the number of cells
in the field of view; (2) detecting the cells in mitotic round-
ing; (3) estimating the general changes in cell activity. Thesethree steps enable extracting measurements for the numericalquantification of photodamage, such as cell size distribution
across time.
D.1. Estimation of the number of cells in the field of view
using virtual cell nuclei staining and segmentation. The to-
tal number of cells in the field of view was computed by
first, inferring the cell nuclei from brightfield images us-
ing an in-house trained Pix2Pix model (20), and then, seg-
menting each nucleus with a pre-trained StarDist model(StarDist-versatile) provided by the developers as "2D versa-tile fluo" model and trained on a subset from the Data ScienceBowl Kaggle Challenge (21). Using ZeroCostDL4Mic (22),
Pix2Pix was trained with paired images of fixed cells stained
with Hoechst and imaged with brightfield and widefield fluo-
rescence microscopy. The imaged data was divided into 400
and132fields of view for training and testing, respectively.
All the input images were preprocessed as follows: a bleach
correction to remove the illumination artefacts was computed
by applying a large low-pass filter (a wide Gaussian filter)
and subtracting it from the original image; then, the image
intensity values were normalised with the min-max projec-tion to the [0,1]range. Both pre-processed input and output
images were then normalised using a percentile normalisa-tion of 1%and99.9% respectively (denoted as Contrast en-hancement in the ZeroCostDL4Mic Pix2Pix notebook). Dur-
ing Pix2Pix training, images are also reshaped to have a
size of 1024◊1024 pixels, so the images were reshaped to
have a pixel size of 0.644 µu◊0.644 µm. Pix2Pix was
trained from scratch with a patch size of 512◊512 pix-
els, a batch size of 5and for 2000 epochs with a learning
rate of 0.001 and 1000 more epochs with a linear learn-
ing rate decay. The accuracy results for the test datasetare as follows: Structural Similarity Index Measure (SSIM)
(0.86), Learned Perceptual Image Patch Similarity (LPIPS)
(0.10), Peak Signal-to-Noise Ratio (PSNR) (25.17), Nor-malised Root Mean Squared Error (NRMSE) (0.13).The out-
put PDF reports of ZeroCostDL4Mic for both the model
training and the quality control check are attached to thesupplementary material. The Pix2Pix model instance corre-
sponding to the last epoch was chosen to process the first
frame of each cell synchronisation video. The output of
Pix2Pix was then segmented using the pre-trained StarDist
model. The total number of individually segmented nucleiwas used as the number of cells in the field of view with an
average of 126cells per FOV .
D.2. Cell mitosis detection with StarDist. A 2D
StarDist (15) model -StarDist-CHO- was trained using
ZeroCostDL4Mic (22) notebooks. From all the videos
acquired for the study, 105 were chosen for the creation
of the ground truth. For each video, a random time-point
was selected and all the rounded cells (either before or
after division) were manually annotated with a unique label(i.e., instance segmentation). The data was split into 85
and 20videos for training and testing respectively. The
model was trained on a Tesla T4 GPU for 500epochs with
patches of 525◊512pixels, a batch size of 20, a grid size
of2◊2for the first convolutional layer of the model, the
Mean Squared Error (MSE) loss function and an initial
learning rate of 5·10
≠3. Data augmentation composed
of rotations, translations and mirroring was also applied.
Before the training, all the images were downsampled to
a pixel size of 0.865 µu◊0.865 µm(i.e., downsampling
of a factor of 1.5709), so the size of a cell in the mitotic
rounding in pixels is smaller than the input size used to train
StarDist. This resampling factor was chosen to ensure thatthe receptive field of the U-Net in StarDist covers a wide
enough region to identify the cell contours characteristic of
the rounded cell. The accuracy results in the test set were as
follows: intersection over union (IoU) (0, 530), false positive
(6.05), true positive (17, 05), false negative (4.6), precision
(0.6712), recall (0.790), accuracy (0.555), f1 score (0, 699),
true detection (21.65), predicted detections (23.1), mean truescore (0.712), mean matched score (0.901), panoptic quality
(0.628). The output PDF reports of ZeroCostDL4Mic for the
model training and the quality control check are attached tothe supplementary material.
D.3. Cell activity estimation. Cell activity was calculated as
the difference between consecutive pre-processed frames.First, a normalisation of the brightfield microscopy videos
was applied to ensure a uniform and comparable illumina-
6 | bioR ‰iv Del Rosario & Gómez-de-Mariscal et al. | PhotoFiTT: Quantifying Live-Cell Phototoxicity. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
E Quantitative data analysis
tion intensity across the entire video: (1) the image intensity
values of each frame were normalised with the min-max pro-jection to the [0,1]range, (2) a bleach correction to remove
the illumination artefacts was computed by applying a largelow-pass filter (a wide Gaussian filter) and subtracting it from
the original image, (3) the image intensity values of the entire
video are normalised again with the min-max projection tothe[0,1]range.Second, the contrast of the video is enhanced
by applying a Contrast-Limited Adaptive-Histogram Equal-ization (CLAHE) using the Skimage Python package with akernel of size 25, a clip limit of 0.01 and256bins. Third, we
applied a Gaussian filter with ‡=1to smooth all the frames
in the video and alleviate the impact of the noise in the im-
ages. Finally, for each time point tin the video, the general
cell activity was computed as the difference between pairs of
temporal frames as follows:
activity (t)=mean!
ÎeIm(t) ≠eIm(t≠1)Î
2"
,(1)
where eIm is a normalised, contrast-enhanced and smoothed
image.
E. Quantitative data analysis. The percentage of cells in
the field of view identified by StarDist-CHO at each time
point t(C(t)) provides a distribution of mitotic rounding
events across time. C(t) is calculated by dividing the number
ofStarDist-CHO detections by the number of nuclei detected
byStarDist-versatile. The peak of cell division is determined
as
tp:C(tp)=max {C(t)} ’tœT, (2)
withtpbeing estimated for each video, V. Cell size was
monitored using the instance segmentations from StarDist-
CHO (Figure 2c-d).
The instance segmentation of StarDist-CHO allows for the
estimation of the cell size (S )(i.e., the sum of all the pixels
forming the cell mask), which is then used to estimate the cell
diameter Dby resolving the equation
S=fiR2, (3)
where Ris the cell radius and D=2R. Tracking the distribu-
tion of Dacross time we distinguish two populations: mother
and daughter cells with a diameter of 20µmand 15µmre-
spectively (Figure 2b,e). This is achieved by classifying all
the detections with a diameter larger than 18µmas mother
cells and daughter cells otherwise (Figure 2f).
The total nnumber of this analysis is the following: for
385nmis50FOVs across 5replicas for unsynchronised
cultures, 50FOVs across 5replicas for 0J/cm2,20FOVs
across 2 replicas for 0.6J/cm2,30FOVs across 3replicas
for6J/cm2and30FOVs across 3replicas for 60J/cm2).
For475nmis 37FOVs across 5replicas for unsynchronised
cultures, 42FOVs across 5replicas for 0J/cm2,30FOVs
across 3 replicas for 0.6J/cm2,27FOVs across 4 replicas
for6J/cm2and14FOVs across 2replicas for 60J/cm2).
For630nmis44FOVs across 5replicas for unsynchronised
cultures, 46FOVs across 5replicas for 0J/cm2,12FOVsacross 2replicas for 0.6J/cm2,44FOVs across 5replicas
for6J/cm2and24FOVs across 3replicas for 60J/cm2).
Cell activity at each time point is hardly difficult to compare
due to the stochasticity of cell movement at short time win-
dows. Therefore, we calculate the cumulative cell activity
afterTminutes given as
cumulative _activity (T)=t=Tÿ
t=0activity (t), (4)
The resulting value at T= 420 min (7h) is shown in
(Figure 2g).
The total nnumber of this analysis is the following: for
385nmis50FOVs across 5replicas for 0J/cm2,20FOVs
across 2replicas for 0.6J/cm2,29FOVs across 3replicas for
6J/cm2and29FOVs across 3replicas for 60J/cm2). For
475nmis36FOVs across 4replicas for 0J/cm2,21FOVs
across 3replicas for 0.6J/cm2,18FOVs across 2replicas
for6J/cm2and4FOVs across 1replicas for 60J/cm2). For
630nmis26FOVs across 3replicas for 0J/cm2,7FOVs
across 2replicas for 0.6J/cm2,22FOVs across 3replicas
for6J/cm2and9FOVs across 1replica for 60J/cm2).
F. SYTOX signal quantification. To assess cellular apop-
tosis, we quantified the percentage of SYTOX-positive cellsby dividing the number of cells expressing SYTOX by the
total cell count in each field of view. The SYTOX signal was
segmented using a two-step process: first, applying Otsu’sthresholding algorithm, followed by a Watershed transforma-tion to delineate individual apoptotic cells. This approach
enabled an accurate approximation of apoptotic cell num-
bers in each frame. The image processing workflow was
implemented in Fiji and automated using a custom ImageJmacro for high-throughput analysis. To determine the totalcell count, we employed the method described in the previ-
ous section, "Estimation of the number of cells in the field
of view using virtual cell nuclei staining and segmentation."
A total of 10FOVs and 1replica for each condition were
assessed.
G. Manual annotations of unsynchronised videos. A
subset of unsynchronised cells was chosen to identify the de-lays in mitosis and cell arrest due to photodamage. Those
cells entering mitosis were manually labelled by detecting the
first time point in which they show the mitotic rounding un-til the resolvable emergence of two daughter cells. Cells thatfailed to produce two daughter cells by the end of the time-lapse (8 hours) but still attained mitotic rounding are consid-
ered to be cell cycle arrested. This annotation lets us distin-
guish between successful mitoses and cell arrest and calcu-late the time for cell division or arrest. A total of 3FOVs per
condition were annotated and tracked.
Del Rosario & Gómez-de-Mariscal et al. | PhotoFiTT: Quantifying Live-Cell Phototoxicity bioR‰iv | 7. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
Note S1: Setting up and using PhotoFiTT in diverse systems
PhotoFiTT is designed to be a versatile and adaptable framework that can be integrated into a wide range of existing live-cell
imaging workflows. Its primary aim is to efficiently evaluate and mitigate phototoxicity. This detailed guide walks researchers
through the steps required for implementing PhotoFiTT, reflecting the components illustrated in Figure 2a.
H. Cell Culture and Synchronisation. The first step in the PhotoFiTT workflow involves preparing the cell samples. This
process can be adapted to various cell types, but for optimal results, adherent cell lines capable of division should be used. The
following steps can be overviewed in Figure 2adiagram blocks: "Unsynchronised cell culture", "Synchronised cell culture",
"Cell synchronisation drug" and "Drug removal".
1.Cell seeding : Seed cells on an appropriate imaging substrate, such as an 8-well chambered cover glass. The seeding
density should be optimised for your specific cell type, but as a starting point, consider using 6◊104cells per well
if using CHO cells. Optimal cell density is crucial for obtaining high-quality images for analysis. Overly dense cellpopulations can complicate image analysis due to overcrowding and may inhibit cells from re-entering interphase after
division due to limited space. Conversely, sparsely populated fields may yield insufficient data for robust statistical
analysis.
2.Synchronisation (optional): Incubate cells with 10nM RO-3306 (a CDK1 inhibitor) for 16≠18hours. This arrests
cells at the G2/M boundary, allowing for a coordinated release into mitosis. It is not advisable to incubate cells for longer
than 20hours in RO-3306 as it can induce apoptosis (1).
3.Unsynchronised option: After seeding, allow cells to grow for the same duration as the synchronised cell population
(16≠18hours).
Recommendation: Conduct simultaneous experiments with both synchronised and unsynchronised cell populations. This
approach allows for the fine-tuning of synchronisation timings and provides insights into the impact of synchronisation on
phototoxicity sensitivity for the specific cell type under study.
I. Photodamage Irradiation. This phase replicates the illumination conditions encountered by cells in fluorescence mi-
croscopy experiments, corresponding to the "Phototoxicity Irradiation" segment depicted in Figure 2a. It’s designed to closely
mimic the intensity and duration of light exposure typical in such experiments, providing a realistic assessment of potentialphototoxic effects.
1.Microscope Configuration: Employ a fluorescence microscope outfitted with light sources that can replicate the specific
illumination conditions required for assessing phototoxicity. This setup is crucial for accurately simulating the light
exposure patterns cells will experience during the experiments.
2.Power Calibration Protocol: Initiate each experimental session by calibrating the photodamage irradiation light’s in-
tensity. This step is paramount to guarantee uniformity in experimental conditions. Employ a power meter to accurately
measure the light intensity (W/cm
2) at the sample plane, ensuring the microscope’s internal chamber has stabilised at
37¶C. Control for the fact that light intensity can vary over time. This calibration is critical before commencing each
experiment to maintain consistency and reliability in results.
3.Illumination Conditions: Configure several fields of view within a single well, assigning unique illumination condi-
tions to each well. This strategy enables the simultaneous evaluation of various light exposure scenarios, facilitating a
comprehensive analysis of phototoxic effects across different intensities and durations.
4.Exposure Process:
•Area Selection: Ensure the selection of distinct, non-overlapping fields for each exposure condition to prevent
double irradiation of the same area. Avoid tiling patterns that might lead to such overlaps.
•Location Tracking: Record the coordinates of each illuminated area. This information is crucial for subsequent
time-lapse imaging, ensuring accurate follow-up on the exposed regions.
•Illumination Application: Implement the predefined illumination protocols precisely for each designated area,
adhering to the established parameters for light intensity and exposure duration.
•Media Change for Synchronised Cells: For cells undergoing synchronisation, wash thoroughly with PBS twice
before media replacement. We use a custom-designed 3D-printed multi-syringe adaptor (available on the PhotoFiTT
GitHub repository) for efficient and uniform media changes across multiple wells simultaneously. This adaptor
is compatible with an 8-well chamber, allowing for simultaneous media changes in 4 wells (Figure S3). The
consistency of the media exchange process is important for accurate comparison of cell division timings acrossdifferent experimental setups. Consider that the interval between synchronisation drug removal and the onset ofmitotic cell rounding is typically under 10minutes.. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
J. Live-cell Imaging. Following the completion of the light irradiation stage, the experimental protocol transitions to the time-
lapse imaging phase. This critical step, denoted as "Acquisition" in the schematic presented in Figure 2a, involves capturing
sequential images over time to monitor and document the cellular responses to the previously applied light conditions.
1.Media Exchange Protocol: Transition the cells to an imaging-optimised medium to enhance both image quality and
cell health during observation. Use phenol-red free Fluorobrite DMEM, fortified with 4mM GlutaMAX for sustained
cellular metabolism, 42µM gentamicin to prevent microbial contamination, and 10% FBS to support cellular growth
and viability.
2.Microscope Configuration:
• Use a high-quality objective that allows to resolve a large number of cells.
• Ensure the imaging chamber is regulated to maintain an environment that mirrors physiological conditions, with a
constant temperature of 37¶Cand a 5%CO2concentration.
3.Acquisition Parameters:
• Use low-intensity brightfield illumination to minimise phototoxicity.• Schedule image acquisition over a span of 6≠8hours, with images captured every 4minutes. This frequency
balances the need for detailed temporal resolution while constraining phototoxicity.
• Select a magnification that allows for clear visualisation of individual cells and their mitotic stages (e.g. a 20X
magnification objective).
4.Multi-Position Imaging Protocol: Configure the microscope for automated imaging of all areas subjected to light
exposure, as well as designated control regions. For experiments incorporating a tiling strategy, align the regions of
photodamage irradiation with those selected for detailed imaging.
Recomentation: Implement adaptive focus control, if available, to ensure stable focus over extended imaging sessions. Cellular
focus can vary, especially between interphase cells and those in mitotic rounding or division. Identify cells at the mitotic
rounding stage to fine-tune the focus, aiming for a balance that accommodates both cell states effectively. This adjustment
should be completed prior to initiating photodamage irradiation and drug removal processes, to prevent any delays in live-cell
imaging that could impact the integrity of the experimental results.
K. Image Analysis Workflow. The PhotoFiTT image analysis pipeline comprises several steps to extract quantitative data
from time-lapse images. This workflow integrates deep learning techniques with traditional image processing methods to
provide a comprehensive analysis of cellular behaviour under various photodamage irradiation conditions. These steps corre-
spond to the workflow blocks "Data annotation", "Identification", "Virtual nuclei staining", "Cell nuclei segmentation", "Mitotic
rounding detection", "Calculation of cell activity" illustrated in Figure 2a.
1.Cell Detection and Quantification (deep learning-based image analysis): We used virtual staining. Alternatively, onecould use existing pipelines or pre-trained models for cell segmentation when available. This processing is only applied
to the first time point of each video.
•Virtual Staining: Apply a Pix2Pix model or an alternative deep learning approach to generate virtual nuclear stains
from brightfield images. This step is crucial for label-free cell detection and is performed only on the first frameof each video sequence. See Methods section B for the experimental acquisition of ground truth data and Methods
section D for the model training. When reproducing our setup, use our pre-trained Pix2Pix model.
•Nuclei Segmentation: Use the pretrained StarDist-versatile model provided by the original authors to segment
individual nuclei in the virtually stained images, providing accurate cell counts and positions.
•Initial Cell Quantification: The number of detected nuclei serves as the baseline cell count for each field of view,
enabling tracking of population dynamics over time.
2.Mitotic Cell Identification (deep learning-based image analysis):
•CHO-specific Detection: For Chinese Hamster Ovary (CHO) cells, employ the specialised StarDist-CHO model
to identify cells in the mitotic rounding phase, leveraging its training on the unique morphology of dividing CHO
cells.
•For other cell lines: For other cell types or experimental conditions, manually annotate a representative image set
and train a new StarDist model following the protocol outlined in Methods Section D.. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
3.Cell Size Analysis and Classification:
•Morphometric Measurements: Exploit the instance segmentations from StarDist-CHO to calculate cell areas and
estimate cell diameters.
•Cell Stage Classification: Categorise cells as "mother" (diameter >18µm) or "daughter" (diameter Æ18µm)
based on their size, enabling tracking of cell division progression.
4.Quantification of Cellular Activity:
•Image Preprocessing: Enhance brightfield images through illumination normalisation, noise reduction, and con-
trast enhancement to optimise subsequent analysis.
•Dynamic Activity Measurement: Compute frame-to-frame differences to quantify overall cellular activity, cap-
turing subtle changes in cell morphology and position.
•Cumulative Activity Assessment: Calculate the total cellular activity over the entire imaging period to assess
long-term effects of light exposure.
Implementation Guidelines:
1.Deep Learning-based Image Processing:
•Local Processing: Install DL4MicEverywhere (2) and execute the appropriate notebooks for Pix2Pix and StarDist
models.
•Cloud-based Processing: Use ZeroCostDL4Mic (3) notebooks on Google Colab for Pix2Pix and StarDist imple-
mentations.
•Pre-trained Models: Access our validated models from Zenodo for immediate use or as starting points for transfer
learning.
2.Data Analysis and Processing:
•Environment Setup: Configure the required Python environment as specified in the PhotoFiTT repository.
•Data Organisation: Structure raw videos and generated masks according to the provided template.
•Analysis Execution: Employ the provided Jupyter Notebooks to process data and replicate the analysis pipeline.
3.Quality Control and Validation:
•Manual Verification: Manually check a subset of images to ensure accurate cell detection and classification.
•Reproducibility Assessment: Compare results across experimental replicates to ensure consistency and reliability.
L. Data Interpretation and Optimisation. The final phase of the PhotoFiTT workflow involves interpreting the results and
using them to refine your imaging protocols. The following analysis is supported by the output of PhotoFiTT’s Jupyter note-
books. It represents the "Data analysis" workflow illustrated in Figure 2a.
1.Mitotic Timing Analysis:
• Plot the distribution of mitotic rounding events over time for each condition.
• Identify the peak division time for control populations (typically around 50minutes post-synchronisation release).
• Quantitatively assess delays in this peak for different light exposures and wavelengths.
2.Cell Size Dynamics Evaluation:
• Monitor the temporal evolution of mother and daughter cell proportions.• Identify delays in daughter cell emergence, indicative of division slowdown.• Identify conditions leading to persistent large cell populations, suggesting potential cell cycle arrest.
3.Cellular Activity Assessment:
• Perform a comparative analysis of cumulative activity levels across different experimental conditions.• Interpret reduced activity as a potential indicator of cellular stress or compromised viability.
4.Phototoxicity Threshold Determination:
• Identify the minimum light dose that induces detectable alterations in mitotic timing, cell size distribution, or
cellular activity.
• Use these thresholds as illumination constraints for designing imaging protocols.. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
Note S2: Optimising Live-Cell Imaging Experiments with PhotoFiTT
To maximise the effectiveness of PhotoFiTT in optimising live-cell imaging experiments, follow these comprehensive steps:
1.Establish a Baseline:
• Conduct initial assessments using a wide range of photodamage irradiation conditions relevant to your typical
experiments.
• This baseline will provide crucial insights into where your current protocols fall on the phototoxicity spectrum.
2.Iterative Protocol Optimisation:
• Use PhotoFiTT results to guide the fine-tuning of acquisition parameters.
• Implement multiple rounds of optimisation as needed, systematically testing new settings.
3.Continuous Monitoring:
• Integrate PhotoFiTT assessments into routine workflow, particularly when:
–Introducing new cell types or modifying culture conditions.
–Incorporating non-standard drugs or chemicals into your experiments.
–Adopting novel fluorophores or labelling strategies.
–Implementing new imaging modalities or upgrading hardware.
4.Informed Experimental Design:
• Leverage PhotoFiTT data to enhance experimental planning:
–Determine the maximum allowable number of time points or z-stacks before phototoxicity becomes a signifi-
cant factor.
–Assess the feasibility of photoactivation or optogenetic experiments based on cellular light sensitivity.
–Optimize the sequence of multi-colour imaging to minimize potential damage from shorter wavelengths.
5.Reporting and Reproducibility:
• Include detailed PhotoFiTT-derived phototoxicity assessments in your methods sections and supplementary data.
This quantitative approach to evaluating phototoxicity will improve the reproducibility and reliability of your imag-ing studies.. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
Fig. S1. Impact of phototoxicity on non-synchronised cell populations. Figure presents the effects of phototoxicity on non-
synchronised adherent mammalian (Chinese Hamster Ovary - CHO) cell populations under near-UV ( 385nm) light exposure. a)
Relationship between light dose and cell fate, showing a decrease in cell division and an increase in cell arrest with higher UV doses. b)Dose-dependent delay in cell division is quantified by measuring the time from mitotic rounding to the emergence of two distinct daughtercells, revealing longer delays at higher near-UV doses. c) Comparison of apoptotic cell rates (indicated by SYTOX-positive staining)
between synchronised and non-synchronised cell populations under near-UV exposure highlights higher vulnerability of synchronized
populations to phototoxic damage.. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
Fig. S2. Cell activity after light exposure. Temporal illustration of near-UV ( 385nm) light dose effects on cell division timing,
arrest, and apoptosis. In the absence or under low light exposure, mother cells (square) proceed to divide resulting in the formation of
daughter cells (square with a cross inside). Following high light exposure, cells present delays in mitotic events resulting in delays or
even apoptosis (triangle). Scale bar 50µm.. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
Fig. S3. 3D-printed multi-well syringe adaptor for efficient media changes. This custom device accommodates four 2.5 mL
syringes for use with standard 8-well glass-bottom imaging chambers. a) 3D render of the .stl file in open and closed configurations.b) 3D-printed device with and without syringes. Black arrows indicate well spacing matching the 8-well chamber; white arrow showsthe plunger thumb rest for uniform syringe motion; orange arrow points to recommended blunt needles for optimal media uptake. c)
Demonstration of the adaptor’s use on an 8-well chamber. The device was designed using TinkerCAD and printed with a Prusa MK4
3D printer. Additional securing with tape is optional.. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
Irradiation range Irradiation average Microscopy modality References
(W/cm2) (W/cm2)
1000 - 20000 10000 STED Wildanger et al., 2008
1000 - 10000 5000 SMLM/RESOLFT Chen et al., 2018;
Grotjohann et al., 2011
100 - 5000 1000 Confocal Icha et al., 2017
50 - 1000 100 SRRF Culley et al., 2018
5 - 100 10 TIRF; SIM Kwakwa et al., 2016;
Li et al., 2015
0.5 - 100 5 LLS; Wide-field Icha et al., 2017;
Schermelleh et al., 2019
Table S1. Irradiation ranges and microscopy modalities for different fluorescence microscopy techniques. This table sum-
marises the typical irradiation ranges and average irradiation intensities for various fluorescence microscopy techniques, along withcorresponding references. STED: Stimulated Emission Depletion; SMLM: Single Molecule Localization Microscopy; RESOLFT: Re-versible Saturable Optical Fluorescence Transitions; SRRF: Super-Resolution Radial Fluctuations; TIRF: Total Internal Reflection Flu-
orescence; SIM: Structured Illumination Microscopy; LLS: Lattice Light-Sheet.. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
Video S1: Detailed procedure for the PhotoFiTT experiment setup and image acquisition. This instructional video
outlines the PhotoFiTT experimental setup and data acquisition process. Initially, standard adherent cells are seeded in suitable
culture dishes. These cells are then synchronised using the CDK1 inhibitor, RO-3306, for a period of 16 to 18 hours to
ensure uniform cell cycle progression. Prior to imaging, the microscope’s excitation light intensity is calibrated to maintain
consistency across all experiments. The cells are exposed to a predetermined dose of excitation light, simulating conditionsencountered during fluorescence microscopy. Following light exposure, the cells are washed with PBS to remove any residual
inhibitor, preparing them for live imaging. The video concludes with the initiation of live-cell acquisition, capturing the
dynamic responses of cells to phototoxic stress.
Video S2. Demonstrating the PhotoFiTT analytical workflow. This video provides a comprehensive walkthrough
of the PhotoFiTT analytical process, utilising the Jupyter notebooks available in the PhotoFiTT GitHub repository(https://github.com/HenriquesLab/PhotoFiTT). The demonstration covers each step in detail, showcasing how to effectively
use the provided tools for analysing phototoxicity effects on cells.
Video S3. Dynamics of adherent cells after light exposure. Synchronised CHO cells following 385 nm (near-UV)
light exposure. Non-exposed cells present a clearly defined peak in mitotic rounding and cell division (white arrow). Cells
exposed to a dose of 0,6J/cm
2present a slight delay but can complete division (orange arrow). Cells exposed to a dose of
6J/cm2present a delay in mitotic rounding. In most cases, the cells present aberrant division (blue arrow) or are unable to
complete division (teal arrow). Cells exposed to the high dose of 60J/cm2become arrested and can result in cell death (purple
arrows).. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint 
----!@#$NewPage!@#$----
Supplementary Bibliography
1. Lyubomir T Vassilev. Cell Cycle Synchronization at the G 2/M Phase Border by Reversible Inhibition of CDK1. Cell Cycle, 5(22):2555–2556, November 2006. ISSN 1538-4101, 1551-4005. doi:
10.4161/cc.5.22.3463 .
2. Iván Hidalgo-Cenalmor, Joanna W Pylvänäinen, Mariana G. Ferreira, Craig T Russell, Alon Saguy, Ignacio Arganda-Carreras, Y oav Shechtman, Guillaume Jacquemet, Ricardo Henriques, et al.
Dl4miceverywhere: deep learning for microscopy made flexible, shareable and reproducible. Nature Methods, pages 1–3, 2024.
3. Lucas von Chamier, Romain F . Laine, Johanna Jukkala, Christoph Spahn, Daniel Krentzel, Elias Nehme, Martina Lerche, Sara Hernández-Pérez, Pieta K. Mattila, Eleni Karinou, Séamus Holden,
Ahmet Can Solak, Alexander Krull, Tim-Oliver Buchholz, Martin L. Jones, Loïc A. Royer, Christophe Leterrier, Y oav Shechtman, Florian Jug, Mike Heilemann, Guillaume Jacquemet, and Ricardo
Henriques. Democratising deep learning for microscopy with ZeroCostDL4Mic. Nature Communications, 12(1):2276, December 2021. ISSN 2041-1723. doi: 10.1038/s41467-021-22518-0 .. CC-BY 4.0 International license available under awas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 16, 2024. ; https://doi.org/10.1101/2024.07.16.603046doi: bioRxiv preprint