Fig 2: Block diagram of IRIF detection. Three-channel images are used for a network input: one for nuclear staining and two for IRIF staining, as exemplified by DAPI staining of the nuclei (blue) and immunodetection of ?H2AX (green) and 53BP1 (red) IRIFs. The process is divided into three steps. First, 3D nucleus masks are created using a U-Net convolutional neural network (CNN) from the channel for nuclear staining. Second, individual IRIFs are detected with the CNN. Third, individual foci are segmented from a multiplied z-stack composed of the two channels for IRIFs by utilizing a maximally stable extremal region (MSER) detector. The output of these three steps is finally merged into nucleus/IRIF 3D masks. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig 3: Block diagram of single-channel detection for ?H2AX and 53BP1 channels in a multichannel image. a. Single-channel detection of IRIFs using the straightforward application of a detection network trained with single-channel manual labels. b. Detection of colocalized IRIFs using labels combined to achieve colocalized labels. The network was trained directly to predict colocalized IRIFs (pre-colocalization). c. Detection of colocalized IRIFs using separated training for individual channels and a combination of results (post-colocalization). d. Boxplot of Dice coefficients achieved using single-channel detection and detection based on pre/post-colocalized IRIFs (b. and c.) on a dataset with individual IRIFs separately annotated for U87/NHDF cells treated with 0.5–8 Gy 30 min/8h PI (679 FOVs).

Fig 4: Zeocin treatment induced more ?H2AX foci and 53BP1 foci in CTLA-4–expressing cells. ?H2AX and 53BP1 protein accumulation at DNA damage sites forms cytologically discernible foci. (A) DNA damage was measured by determining the number of ?H2AX and 53BP1 foci using immunofluorescence detection. Representative images are presented. The B7-1/B7-2-/- and B7-1/B7-2/CTLA-4-/- T cells were treated without or with 500 µg/ml zeocin for 1 h followed by culture in a normal medium for the indicated times and then stained for ?H2AX (red), 53BP1 (red), and nuclei (DAPI, blue). Scale bar, 20 µm. (B) Quantitation of the data shown in (A). The histograms represent the percentage of cells with more than 10 ?H2AX and 53BP1 nuclear foci. Each experiment was independently performed three times (n = 3), and more than 100 cells were counted in each experiment. The data are presented as the mean ±SEM; ***p < 0.001.

Fig 5: Mild replication stress induces whole chromosome mis-segregation. a Mean percentage of ?H2AX and 53BP1 positive RPE1 cells. N = 3 independent experiments examining NT = 909, 200 nM Aph = 963, 400 nM Aph = 971 and 1 mM HU = 637 cells; p < 0.0001 in two-way Anova test (yH2AX: NT vs. 400 nM Aph p < 0.0001; NT vs. HU p < 0.0001; 53BP1: NT vs. HU p < 0.0001). b Representatives images of cells stained for ?H2AX and DNA after indicated treatments. All scale bars in Figure = 10 µm. c Number of metaphase chromosome breaks. N = 3 examining NT = 222 and Aph = 212 metaphase spreads; p = 0.0008 in one-way Anova test (NT vs. 200 nM Aph p = 0.0114; NT vs. 400 nM Aph p = 0.0005). d Representative image of a Giemsa stained metaphase after a 400 nM Aph treatment; insets show an intact and broken chromosome. e Representative images of RPE1 cells stained for 53BP1 and DNA after indicated treatments. f Percentage of chromosome segregation errors as quantified from time-lapse imaging experiments of RPE1 H2B-mCherry/EB3-GFP cells. N = 3 (Aph) and 6 (NT) examining NT = 551, 200 nM Aph = 423 and 400 nM Aph = 195 anaphases; p = 0.0003 in two-way Anova test (lagging DNA: NT vs. 200 nM Aph p = 0.04; NT vs. 400 nM Aph p = 0.0001). g Representative time-lapse sequence after indicated treatments quantified in (f). Nuclear envelope breakdown (NEBD) is t = 0. White arrow indicates lagging DNA. h Conversion rate of lagging DNA to micronuclei according to the left schematic (created by TW). i Representative time-lapse sequence of cells forming a micronucleus from persistent lagging DNA (white arrow). NEBD is t = 0. j Percentage of interphase cells containing CENP-A-negative and CENP-A-positive micronuclei after indicated treatments. N = 3 examining NT = 2928, Aph = 2421 and Monastrol = 2653. CENP-A positive MN: NT vs. Aph p = 0.0269 in unpaired t-test; NT vs monastrol p = 0.0029. k Representatives image of cells with micronuclei containing (right) or not (left) CENP-A signal; error bars indicate sem. Source data are provided as a Source Data file