To enable comparison of cell survival after low-LET X-ray exposure and high-LET $\alpha$-particle exposure (see Supplementary material, Supplementary Figs. 1,2, and Supplementary Tables 1,2), a custom cell container called “ring system” was used for cell survival experiments. This was due to the high shielding effect of $\alpha$-particles by the plastic and glass bottoms of common cell culture equipment. In this system, cells were seeded on a 4.7 µm thin polypropylene foil, which was clamped between two stainless steel rings (Ø 3.8 cm), resulting in a growth area of 10 cm². For increased cell attachment, the foil was coated with gelatin (Sigma-Aldrich, St. Louis, MO, USA, G1890) before cell seeding. The coating was performed using a 0.1% gelatin solution dissolved in sterile water, with an incubation time of 15 minutes at 37 °C. Per ring, 250,000 cells (25,000 cells/cm²) were seeded one day before the experiment. The successful penetration of $\alpha$-particles through the foil and the gelatin coating was verified by a focus immunofluorescence assay before the experiments (see Supplementary Fig. 1).

Cell irradiation with low-LET X-rays was performed using a CellRad X-ray irradiation system (Precision X-Ray Inc., Madison, WI, USA) at a dose rate of 1.6 Gy/min (5 mA, 130 kV, 0.5 mm aluminium filter). The sample had to be placed centrally on the detector in the X-ray cabinet so that all cells in the ring system were irradiated and no shadowing effects occurred due to the high walls of the ring system. To track the exact dose irradiated during the cell survival experiments, dosimetry measurements were performed using XR-RV3 Gafchromic™ films (Ashland Inc., Wilmington, DE, USA, 832486) for each sample. For each dose (Sham, 1 Gy, 2 Gy, 3.7 Gy, 5.2 Gy, and 7.2 Gy), five to six samples were irradiated in the cell survival experiments.

Cell survival was determined by performing CFA as established in Wank et al. [29]. After irradiation, the cells were detached from the foil using trypsin (Sigma-Aldrich, St. Louis, MO, USA, T4299-100) and counted twice in a Fuchs-Rosenthal C-Chip (NanoEnTek, Seoul, Korea, DHC-F01). A total of 284 to 1198 cells were counted per sample. The cell suspension was diluted to the desired cell density (see Table 1) and seeded into 12-well plates (Greiner Bio-One GmbH, Frickenhausen, Germany, 665180). Per sample, three 12-well plates were seeded. After seeding, the samples were incubated for 12 days at 37 °C, 5% CO₂, and 95% humidity. At the end of this incubation period, the cells were fixed with ice-cold methanol (Carl Roth GmbH + Co. KG, Karlsruhe, Germany, CP43.3) and stained with a 0.1% (w/v) crystal violet solution (Sigma-Aldrich, St. Louis, MO, USA, C0775). The plates were scanned with a GelCount™ platform (Oxford Optronix Ltd., Adderbury, UK), and the numbers of colonies were counted by hand. A colony was counted as one if it consisted of $\ge$50 cells.

The plating efficiency (PE) was determined for each sample. PE is defined as the percentage of seeded cells that grow to a colony:

$$PE=\frac{number of counted colonies}{number of seeded cells}.$$

The survival fraction (SF) of cells treated with a given dose was calculated by dividing the mean PE_i⁢r⁢r of the irradiated samples by the mean PE_{s⁢h⁢a⁢m} of the sham-irradiated samples (control):

$$SF=\frac{P{E}_{irr}}{P{E}_{sham}}.$$

The uncertainties of the SF values were calculated by Gaussian error propagation, and finally, the survival curves were fitted by the linear-quadratic (LQ) model:

$$SF\left(D\right)=exp\mathrm{}(-\alpha \cdot D-\beta \cdot D^2)$$

where D is the dose, $\alpha$ and $\beta$ are free fit parameters corresponding to the linear and quadratic radiation responses, respectively.

Two independent experiments were performed. The first experiment yielded a mean PE $\pm$ standard deviation of (0.36 $\pm$ 0.04) % for the sham-irradiated samples. The second experiment yielded a mean PE of (0.499 $\pm$ 0.026) %. According to one-way ANOVA analysis, the two experiments were equivalent, so the samples were pooled for analysis.

For TNT network analysis, the cells were seeded in a glass-bottom CELLview cell culture dish 35 $\times$ 10 mm with four wells (Greiner Bio-One GmbH, Frickenhausen, Germany, 627870) at a seeding density of 55,000 cells/mL (1 mL volume per well; 28,947 cells/cm²) one day before the experiment.

Before X-ray irradiation, the cellular membrane was stained with a 1.5X CellMask™ Plasma Membrane staining solution (Thermo Fisher Scientific Inc., Waltham, MA, USA, C3760 for green, C10045 for orange) dissolved in growth medium. After 10 minutes of incubation at 37 °C, 95% humidity, and 5% CO₂, the cells were washed three times with growth medium, followed by imaging in growth medium. The green dye was used for co-staining experiments, and the orange dye was used for acquiring live-cell videos for TNT network analysis, as this dye is more photostable [30].

Cell irradiation with low-LET X-rays was performed using the same CellRad X-ray irradiation system (Precision X-Ray Inc., Madison, WI, USA) at a dose rate of 1.6 Gy/min (5 mA, 130 kV, 0.5 mm aluminium filter) as for the cell survival experiments. For each experiment, one four-well dish was used. To enable direct comparison of samples treated with different doses in TNT-network experiments, two wells were irradiated with 1.8 Gy (corresponding to 50% cell survival), and the other two wells were irradiated with 3.9 Gy (corresponding to approximately 10% cell survival). A tungsten shield ensured that each well received only its target dose. The experiment was performed in triplicate, resulting in six samples per dose (two samples per dose per experiment). Since dose measurements using XR-RV3 Gafchromic™ films with the 5 mm thick tungsten shield used resulted in a radiation dose penetration of (4.2 $\pm$ 0.6; mean $\pm$ SD) %, the sham-irradiated controls were performed separately. A total of eight samples (two experiments with one four-well dish each, i.e., four samples) were analyzed for sham irradiation.

Confocal live-cell imaging was performed using a Leica TCS SP8 3X confocal microscope (Leica Microsystems CMS GmbH, Mannheim, Germany) equipped with a live-cell imaging unit consisting of a climate box and CO₂ supply with a humidifier (“The Cube & the Box”, Life Imaging Services GmbH, Basel, Switzerland). Cells were imaged under live-cell conditions of 37 °C and 5% CO₂. For imaging, a 63x oil objective (Leica Microsystems CMS GmbH, Mannheim, Germany) with a high numerical aperture of 1.4 and the halogen- and fluorescence-free immersion oil Immersol™ 518 F/37 °C from Zeiss (Carl Zeiss Microscopy Deutschland GmbH, Oberkochen, Germany, 444970-9010-000) with a refractive index of 1.518 at 37 °C were used. The power of the white light laser (WLL) was kept as low as possible during the live-cell imaging to reduce cell stress. The laser power used for excitation was approximately 1 mW. A scanning speed of 600 Hz was chosen. Additionally, the cells were scanned bidirectionally to reduce motion artifacts and cell stress due to long light exposures. The fluorescence signal was detected using a hybrid detector (HyD, Leica Microsystems CMS GmbH, Mannheim, Germany), which combines the sensitivity of a photomultiplier with the noise suppression of an avalanche photodiode.

For the TNT network analysis, cell videos were acquired up to 24 hours after irradiation. The time needed to prepare and check the microscope settings for acquiring videos immediately after cell irradiation was minimized. Defining the areas to be imaged, creating focus maps for each area, and setting the remaining parameters took a total of less than 50 minutes. Between the end of the first irradiation and the start of imaging, a mean time $\pm$ standard deviation of (37 $\pm$ 6) minutes elapsed. Using dishes with four separate wells allowed us to image four different treatments simultaneously, ensuring ideal comparability. One area was imaged per well.

A 3 $\times$ 3 mosaic image with 10% overlap was acquired per well, resulting in a final merged field of view of approximately 510 µm $\times$ 515 µm. At the beginning of the video, the field of view contained a mean of 43 $\pm$ 15 LN229 cells and 53 $\pm$ 11 U87 MG cells for studying the effect of X-ray radiation on TNTs. All videos were acquired as 20-step z-stacks with a step size of 400 nm, resulting in an acquired height of 7.6 µm. The pixel size was 80 nm. The live-cell videos are 24 hours long, with a time interval of 30 minutes. To keep the cells in focus throughout the videos, a 9-point rectangular focus map was defined for each area, and the autofocus feature of the microscope’s LAS X software (Leica Application Suite X, Leica Microsystems CMS GmbH, Mannheim, Germany, version 3.5.6) was used. For the first two to three time points, a WLL intensity of 10% was sufficient for a good signal; thereafter, the intensity was increased to 15%. The excitation wavelength used and the detector range used for the membrane labeling (CellMask™ Orange Plasma Membrane Stain) are listed in Table 2.

We identified membrane connections as TNTs if they connected two separate cells and were smaller than 1 µm in diameter, and if they were located above the substrate. This criterion was determined by acquiring z-stacks for imaging and evaluating TNTs using 3D images rather than maximum projections. A previous study [30] thoroughly examined the characteristics of TNTs in U87 MG and LN229 cells, including their morphology, lifetime, formation, dimensions, and the optimized methodology for studying TNTs using suitable confocal live-cell imaging.

To capture potential communication phases during recovery after irradiation, TNT networks were evaluated at multiple time points (1, 3, 10, and 24 hours), as described by Matejka and Reindl [28]. TNTs were manually counted by scrolling through the z-stack image and examining each cell individually. Cells that were connected via TNTs were referred to as “connected cells”. Depending on the number of TNTs establishing the connection between the two cells, the connection was referred to as either a “simple” or a “complex connection”. Connections consisting of one or two TNTs were referred to as simple connections, and connections consisting of three or more TNTs were referred to as complex connections. Two reasons were given for this distinction. First, as reported by Matejka and Reindl [28], the individual TNTs of very dense connections consisting of five or more TNTs were so closely packed that they were indistinguishable and therefore not countable for evaluation purposes. Second, the number of TNTs between two cells can indicate the strength or efficiency of a connection because the more TNTs there are between two cells, the more cargo can be transported at the same time. According to our definition, then, a complex connection could be more efficient than a simple one. Further subclassification by TNT number alone was not recommended because connections containing exactly two, three, or four TNTs are rare.

Cells connected to at least two other cells were classified as “highly connected cells” and used as a proxy for network complexity. For each imaging field, the total number of cells was recorded, and the TNT network was quantified by the following: (i) the percentage of connected cells, (ii) connections per cell (total connections divided by cell count, irrespective of connection type), (iii) the percentage of complex connections, and (iv) the percentage of highly connected cells. Expressing the TNT metrics as percentages (or dividing by the total cell count) normalized the results for differences in cell density between fields. Finally, the irradiated samples were compared with the sham-irradiated controls.

For co-staining experiments, the cells were seeded in glass-bottom dishes (µ-Dish 35 mm, Ibidi GmbH, Gräfelfing, Germany, 81158) at a seeding density of 65,000 cells/mL (2 mL total volume; 31,707 cells/cm²) one day before the experiment. All glass-bottom products used had a glass thickness of (170 $\pm$ 15) µm and were, therefore, suitable for high-resolution microscopy.

For calcium staining, the BioTracker™ 609 Red Ca²⁺ AM dye (Sigma-Aldrich, St. Louis, MO, USA, SCT021) was used. The fluorescence intensity of this dye changes reversibly depending on the Calcium (Ca²⁺) concentration. Cell labeling was performed at a dye concentration of 10 µM with the addition of 0.05% Pluronic F-127 (Thermo Fisher Scientific Inc, Waltham, MA, USA, P6866), a nonionic surfactant polyol that inhibits dye aggregation, and an incubation time of 1 hour to 1.5 hours at 37 °C, 95% humidity, and 5% CO₂. A serum-free medium was used as the loading medium, as labeling in normal growth medium failed, and staining in Hanks’ Balanced Salt Solution (HBSS) for 1 hour led to cell death. An incubation time of 30 minutes resulted in a weaker signal, so the incubation time was increased to at least 1 hour.

Mitochondria were stained with Tetramethylrhodamine ethyl ester (Enzo Biochem Inc., Farmingdale, NY, USA, ENZ-52309), tetramethylrhodamine ethyl ester (TMRE) for short. Cells were stained with a 50 nM staining solution with growth medium for 25 minutes at 37 °C, 95% humidity, and 5% CO₂. TMRE is a cell-permeable, cationic, red-orange fluorescent dye that accumulates at the negative membrane potential in active mitochondria, but is not cytotoxic [31]. It is therefore highly selective for functional mitochondria and produces a very low background signal in living cells. Non-functional mitochondria that have been depolarised (e.g., by radiation) are not stained by this dye. Due to its ability to stain intact mitochondria specifically and its stable labeling of all cells with a superior signal-to-noise ratio compared to MitoTracker™ Green and MT-mKeima-Red in HeLa cells, TMRE was successfully used in previous experiments investigating mitochondrial depolarization by radiation in our laboratory [32] and was therefore also selected for this study.

The same confocal system, scanning parameters, and objective were used for the co-staining experiment as for the TNT network analysis. For imaging, a system-optimized pixel size of 76 nm and a z-step size of 299 nm were chosen. Confocal microscopy of living cells co-stained with CellMask™ Green Plasma Membrane Stain and BioTracker™ 609 Red Ca²⁺ AM dye was performed using two hybrid detectors (HyDs) simultaneously. The first HyD detected the plasma membrane signal (CellMask™ Green Plasma Membrane Stain), and the second HyD detected the signal of Ca²⁺ (BioTracker™ 609 Red Ca²⁺ AM dye). The selected detector ranges and the excitation wavelengths used are listed in Table 2.

For imaging living cells co-stained with the CellMask™ Green Plasma Membrane Stain and TMRE, sequential microscopy was performed because the fluorescence spectra of the two dyes used are very close. The signal of the membrane dye was recorded in the first sequence, and the signal of TMRE was recorded in the second sequence. The excitation wavelengths and detector ranges used are listed in Table 2. The chosen excitation wavelength and detector range for the TMRE signal are not optimal because they do not cover the maxima. However, these values were chosen to minimize overlap and interference between the two signals. To ensure rapid acquisition and to minimize cell movement during acquisition, the sequence was switched between stacks.

The analysis of the cargo transport inside TNTs was performed using the raw images. For image visualization, however, the images were deconvolved using Huygens Professional Deconvolution software (Scientific Volume Imaging B.V., Hilversum, Netherlands, version 23.04). Based on the microscopy parameters (metadata) used, the software calculated a theoretical point spread function (PSF). Deconvolution was carried out using the Classic Maximum Likelihood Estimation (CMLE) iterative algorithm, which was established in our lab for several approaches. The deconvolution process was performed with a signal-to-noise ratio (SNR) of three to five, an automatic background estimation (estimation mode: lowest), and an area radius of 0.7 µm.

Glioblastoma patients are usually treated with photon radiotherapy. To understand how cellular communication along TNTs may affect therapeutic outcome, the response of TNT networks to X-ray irradiation was investigated. For this purpose, U87 MG and LN229 cells were imaged using confocal microscopy up to 24 hours after X-ray irradiation. Four-well cell culture dishes were used for the experiments, allowing live cell microscopy of up to four samples in parallel and ensuring good comparability. Before irradiation, the cell membrane was labeled using the CellMask™ Orange plasma membrane stain to visualize the TNTs. Cells were irradiated at the following doses: 1.8 Gy, corresponding to 50% cell survival, and 3.9 Gy, corresponding to approx. 10% cell survival. The TNT networks were analyzed from the obtained cell videos according to their expansion, i.e., how many cells were connected by TNTs and how many connections exist per cell, and their complexity, i.e., how many complex connections consisting of three or more TNTs and how many highly connected cells (cells connected to at least two other cells) are present in the TNT networks. These parameters were determined at 1 hour, 3 hours, 10 hours, and 24 hours after irradiation to capture possible communication phases during the recovery phase after irradiation. Finally, the irradiated samples were compared with sham-irradiated controls.

For the TNT network connectivity, the number of connected cells as well as the number of connections per cell were considered. Fig. 2 shows the results of TNT network expansion in LN229 ((a) and (c)) and U87 MG cells ((b) and (d)) after sham (black), 1.8 Gy (red), and 3.9 Gy (blue) X-ray irradiation. In (a) and (b), the time course of the fraction of connected cells is shown, and in (b) and (c), the time course of the number of connections per cell is shown. An increased TNT network expansion is measurable in both cell lines after 1.8 Gy X-ray irradiation.

Considering the LN229 cell line, significantly (p = 0.0264) more LN229 cells were connected by TNTs in the 1.8 Gy irradiated samples compared to the sham irradiated controls ((52 $\pm$ 5) % vs. (36 $\pm$ 4) %; (mean $\pm$ SEM)) 10 hours after irradiation. This significant increase in TNT network expansion is pronounced with time as the fraction of connected cells increases to a value of (54.1 $\pm$ 2.8) % in LN229 samples irradiated with a dose of 1.8 Gy, which is 1.3 times higher (two-sample t-test, p = 0.0049) compared to (40.8 $\pm$ 2.6) % connected cells in the shams at the 24 hours time point.

The increased TNT network expansion was also measurable when considering the number of connections per cell. In LN229 cells, a significant increase (two-sample t-test, p = 0.0355) between the 1.8 Gy X-ray exposure and sham control was already visible 3 hours after exposure. In the shams, approximately every 6th cell has a connection (i.e., 0.175 $\pm$ 0.027 connections per cell), whereas in the irradiated samples, every 4th cell has a connection (i.e., 0.258 $\pm$ 0.018 connections per cell). This significant increase is again pronounced with increasing time (two-sample t-tests; 10 hours: p = 0.0063; 24 hours: p = 0.0098). At these times, approximately 1.5 times more connections per cell were detected in 1.8 Gy irradiated samples (10 hours: 0.38 $\pm$ 0.04; 24 hours: 0.40 $\pm$ 0.029) compared to the connections per cell in sham-irradiated controls (10 hours: 0.229 $\pm$ 0.026; 24 hours: 0.28 $\pm$ 0.026) corresponding to almost every second cell with connection in the irradiated and barely every 4th cell in the sham.

In U87 MG cells, a significant (two-sample t-test, p = 0.0459) 1.2-fold higher number of connections per cell was detected 10 hours after 1.8 Gy irradiation compared to sham (1.10 $\pm$ 0.08 vs. 0.90 $\pm$ 0.05). Regarding the proportion of connected cells, there is a trend of increased connectivity 10 hours after 1.8 Gy X-ray irradiation, with (86.9 $\pm$ 2.8) % of cells connected by TNTs compared to (81 $\pm$ 4) % in the sham-irradiated cells. The difference became significant (two-sample t-test, p = 0.0163) 24 hours after irradiation: (91.3 $\pm$ 1.8) % of the cells in the irradiated populations were connected, compared to (83.5 $\pm$ 2.0) % of the cells in the sham-irradiated populations.

For the higher dose of 3.9 Gy, no significant differences in TNT network expansion were observed for either cell line. However, the measured values were slightly higher than those for the sham exposure in both cell lines. Supplementary Figs. 3,4 show some representative images of TNT networks 10 hours after irradiation in LN229 cells and U87 MG, respectively.

The TNT network complexity was quantified by the number of highly connected cells and the proportion of complex connections. Considering the number of highly connected cells, the significant increase in TNT network expansion at about 10 hours after 1.8 Gy irradiation also resulted in a higher proportion of highly connected cells in both cell lines (see Fig. 3a,b, respectively), and thus in a more complex TNT network. At 10 hours and 24 hours after irradiation, there were almost twice as many highly connected cells in LN229 samples irradiated at a dose of 1.8 Gy (10 hours: (15.4 $\pm$ 2.9) %; 24 hours: (19.2 $\pm$ 2.5) %) compared to sham-irradiated samples (10 hours: (8 $\pm$ 1.3) %; 24 hours: (11.9 $\pm$ 1.7) %), resulting in a significant increase (two-sample t-tests; 10 hours: p = 0.0255; 24 hours: p = 0.0276) in the complexity of the TNT networks. In U87 MG cells, there was a significant (p = 0.0248) increase in the proportion of highly connected cells between 1.8 Gy X-ray irradiation with a value of (62.6 $\pm$ 2.4) % and sham irradiation with a value of (53.7 $\pm$ 2.4) % 10 hours after irradiation. This increase was attenuated 24 hours after irradiation.

The TNT networks of LN229 cells did not show increased complexity with respect to the proportion of complex connections. At the beginning of the cell videos (1 hour and 3 hours after irradiation), high variations (e.g., ranging from 16% to 56% within the error bars at 1) in the proportion of complex connections were measured (see Fig. 3c). These large variations are probably due to the low overall number of TNTs detected in the LN229 cells (2–20 TNTs per sample). The variations became smaller at 10 hours and 24 hours after irradiation (e.g., ranging from 20.2% to 32% within the error bars at 24 hours). At these times, more TNTs were detected (see results on connections per cell shown in Fig. 2c), and therefore, a better statistic regarding the proportion of complex connections was achieved. The proportion of complex connections in LN229 cells seemed to decrease slightly over time.

In U87 MG cells, the proportion of complex connections remained relatively constant over time, at approximately 25% in sham-irradiated controls (see Fig. 3d). In contrast, TNT networks in irradiated cells tend to have slightly more complex connections over time. Starting with (25 $\pm$ 4) % for 1.8 Gy irradiation and (27 $\pm$ 5) % for 3.9 Gy irradiation at 1 hour after irradiation, the proportions increased 24 hours after irradiation to (34.1 $\pm$ 1.9) % and (33.7 $\pm$ 1.8) %, respectively. This is a significant increase over sham for both doses, 1.8 Gy (two-sample t-test, p = 0.0168) and 3.9 Gy (two-sample t-test, p = 0.0199). The proportions of complex connections were approximately 1.3 times higher in the irradiated samples compared to the sham, with a value of (25.4 $\pm$ 2.3) % at 24 hours after irradiation.

In addition to performing a statistical analysis using two-sample t-tests, two-way ANOVAs with repeated measures were conducted to identify significant differences between groups (i.e., the doses). These analyses revealed no significant differences between sham irradiation, 1.8 Gy, and 3.9 Gy X-ray irradiation in U87 MG cells. However, significant differences between the groups (i.e., the doses) were found for LN229 cells in terms of the number of connected cells, the number of connections per cell, and the number of highly connected cells. Subsequent Bonferroni post hoc tests revealed significant differences when comparing sham irradiation and 1.8 Gy X-ray irradiation, with p-values of 0.005, 0.002, and 0.033, respectively, for the number of connected cells, connections per cell, and highly connected cells. According to the two-way ANOVAs with repeated measures, no interactions between treatment and time were found in both cell lines.