Stochastic asymmetric repartition of lytic machinery in dividing human CD8+ T cells generates heterogeneous killing behavior

Cytotoxic immune cells are endowed with a high degree of heterogeneity in their lytic function, but how this heterogeneity is generated is still an open question. We therefore investigated if human CD8+ T cells could segregate their lytic components during telophase, using imaging flow cytometry, confocal microscopy and live cell imaging. We show that CD107a+-intracellular vesicles, perforin and granzyme B unevenly segregate in a constant fraction of telophasic CD8+ T cells during each division round. Mathematical modeling posits that unequal lytic molecule inheritance by daughter cells results from the random distribution of lytic granules on the two sides of the cleavage furrow. Finally, we establish that the level of lytic compartment in individual CTL dictates CTL killing capacity. Together, our results show the stochastic asymmetric distribution of effector molecules in dividing CD8+ T cells. They propose uneven mitotic repartition of pre-packaged lytic components as a mechanism generating non-hereditary functional heterogeneity in cytotoxic cells.

human CTL belonging to the same clonal population exhibit heterogeneity in their lytic 59 function during sustained interaction with target cells (Vasconcelos et al., 2015). While, 60 some CTL kill a limited number of target cells, others emerge as super-killer cells. 61 One proposed mechanism of functional heterogeneity generation in T lymphocytes is 62 asymmetric cell division (ACD). ACD is a key mechanism to generate cell heterogeneity 63 in biology. It plays a crucial role in embryogenesis by allowing the formation of two 64 distinct cells from a single mother cell (Dewey et al., 2015;Knoblich, 2008). In 65 immunology, ACD has been proposed as a process allowing mouse naive T lymphocytes 66 to divide into short-lived effector T cells and memory T cells, after TCR-triggered division 67 (Arsenio et al., 2015(Arsenio et al., , 2014Chang et al., 2011Chang et al., , 2007. 68 In the present work, we investigated the possibility that, in dividing human CD8 + T cells, 69 heterogeneous distribution of molecules relevant for cytotoxic function into nascent 70 daughter cells might contribute to CTL killing heterogeneity. 71 6 marker of cell division (allowing us to identify cells in the different division rounds (Quah 118 and Parish, 2012)), and to define total protein repartition in telophase (Filby et al., 2011). 119 This procedure minimized the possibility that, if some images were taken slightly on an 120 angle, with one daughter cell slightly more in focus than the other, the markers of interest 121 would artificially appear as asymmetric. Indeed, asymmetric distribution was defined as 122 cells in telophase in which repartition of the marker of interest in the nascent daughter cells 123 was beyond the 40-60 % limits observed for CTV repartition (Figure S2 B). In addition, 124 to further exclude the possibility of measurement artifacts, we verified individual cells by 125 eyes and included in the analysis only cells in telophase that were on a even plane. 126 Specificity of staining for the various markers was validated (see Material and Methods 127 section). 128 In a first approach, CD8 + T cells freshly isolated from healthy donor blood samples were 129 stimulated with immobilized anti-CD3/anti-CD28/ICAM-1 for 72 hours. Anti-CD3/anti-130 CD28/ICAM-1 stimulation resulted in activation of human CD8 + T cells as shown by cell 131 proliferation and CD137 up-regulation ( Figure S3). Repartition of the lysosomal marker 132 CD107a was investigated in cells in telophase. As shown in Figure 1A, while CTV 133 distribution ranged between 40-60% in dividing T cells, 23 % of telophasic CD8 + T cells 134 exhibited an uneven distribution of CD107a + vesicles overcoming the 40-60% CTV range. 135 We next investigated the distribution in telophase of lytic components such as perforin and 136 granzyme B (GrzB), molecules known to be pre-stored in lytic granules. As shown in 137 The slope of the linear regression curve for the distribution of CD107a, perforin and GrzB 141 as compared to CTV was close to 0.1, indicating that these 3 molecules distributed 142 independently from total proteins. 143 To define whether uneven repartition of lytic components could be observed in fully 144 differentiated cells, such as memory cells, we investigated CD107a and perforin 145 distribution in telophase in purified human memory CD8 + T cells. This analysis showed 146 that also memory CD8 + T cells exhibited uneven repartition of CD107a and perforin in 147 telophase ( Figure S4). 148 We next investigated whether lytic machinery asymmetric repartition could also be 149 observed in activated CD8 + T cell populations composed of monoclonal cells such as 150 antigen-specific CTL clones. To address this question, we investigated CD107a repartition 151 in CTL undergoing cell division. For this study, we activated CTL clones using 152 immobilized anti-CD3/anti-CD28/ICAM-1 for 72 hours. We opted for this stimulation 153 condition since, in preparatory experiments, we observed that conjugation of CTL with 154 cognate target cells, results (during the 72 hours culture) in the creation of cellular clumps 155 and debris due to CTL killing activity, thus making it difficult and potentially misleading 156 to analyze cells by image flow cytometry and conventional microscopy. As shown in 157 Figure 1D, we observed that in clonal CTL undergoing cell division, 15% of the two 158 nascent daughter cells in telophase exhibited uneven distribution of CD107a, thus 159 confirming and extending observations obtained using CD8 + peripheral blood T cells. 160 Taken together, the above results indicate that a lysosomal-associated membrane protein 161 known to be a marker of lytic granules and effector molecules involved in CTL lytic 162 function, unevenly segregate in 10-23 % of individual human CD8 + T cells undergoing Image flow cytometry allows the unambiguously identification and capture of rare events 168 within a cell population, such as cells in telophase, albeit exhibiting a lower resolution 169 when compared to classical imaging methods. This notion prompted us to confirm results 170 obtained using imaging flow cytometry, with additional methods. 171 We therefore used 3D confocal laser scanning microscopy to measure CD107a content in 172 telophasic CD8 + T cells following stimulation with immobilized anti-CD3/anti-173 CD28/ICAM-1. Although this approach allowed us to collect a relatively small number of 174 cells in telophase (n=61 compared to n=908 obtained by image flow cytometry), it revealed 175 that 27% of the CD8 + T cells in telophase exhibited uneven repartition of CD107a, above 176 a 1.5 threshold (corresponding to the 40-60% ranged used in imaging flow cytometry 177 experiments) (Figure 2A). Figure  Together, the above results indicate that confocal laser scanning microscopy provides 183 results that reinforce those we obtained using imaging flow cytometry and supports the 184 finding that lytic granules undergo uneven repartition in ~20% of dividing CD8 + T cells. anti-CD3/anti-CD28/ICAM-1 for 72 hours. As shown in Figure 3A and B, both T-bet and 199 c-myc did not unevenly segregate into the two nascent daughter cells during telophase. 200 Moreover, the slope of the linear regression curve for the distribution of T-bet and c-myc 201 as compared to CTV was close to 1, indicating that the repartition of these 2 molecules in 202 telophase followed that of total proteins. 203 To further define whether the observed uneven repartition of lytic components was or was 204 not related to ACD, we investigated whether uneven repartition of lytic components was 205 dependent on a polarity cue (e.g. localized TCR stimulation) as previously described for 206 ACD (Arsenio et al., 2015;Pham et al., 2014). Figure 4A and B shows that a polarity cue 207 was not required to induce uneven distribution of lytic molecules, since comparable 208 CD107a + vesicles segregation was observed in peripheral blood CD8 + T cells stimulated 209 by either immobilized (anti-CD3/anti-CD28/ICAM-1) or soluble (PMA + ionomycin) 210

stimuli. 211
Overall, the above results demonstrate that uneven partitioning of lytic compartment in 212 telophase is not associated with asymmetric segregation of fate determining transcription 213 factors. Moreover, a polarity cue is not required. All in all, the above results show that, in 214 human CD8 + T cells, lytic machinery uneven repartition is not related to described We next investigated whether lytic machinery uneven repartition occurred during 220 subsequent divisions and whether this process could be involved in preserving lytic 221 machinery heterogeneity within CD8 + T cell populations. 222 We considered the cells in the different rounds of division (identified by different peaks of 223 CTV dilution, Figure S3) and analyzed CD107a repartition in telophasic cells. This 224 analysis showed that, in all division rounds considered, a comparable percentage of cells 225 underwent heterogeneous repartition of CD107a ( Figure 5A and B). 226 A complementary observation indicated that the heterogeneity process is stationary but not 227 hereditary: e.g. a daughter cell originating from a heterogeneous division has a constant 228 stationary probability to produce a new uneven division. We arrived to this conclusion by 229 generating CD107a fluorescence intensity (CD107a-FI) density curves of all telophasic 230 cells having undergone 0, 1 or 2 mitosis. Cells in telophase showing unequal CD107a-FI 231 repartition were then plotted on these curves ( Figure 5C). The  2 statistical test showed 232 that these cells were randomly and independently distributed on the CD107a-FI density 233 curves, supporting the hypothesis that there is no inheritance in the decision to divide 234

unevenly (see Materials and Methods section). 235
We next asked whether this process might create a drift in lytic compartment content in 236 daughter cells leading to the emergence of cellular subsets expressing higher or lower 237 levels of CD107a. To address this question, we analyzed the total CD107a-FI in all G1 238 cells (either undivided or following each division round). As shown in Figure 5D, the total 239 CD107a-FI appeared to be broadly similar in the different rounds of division in the whole 240 populations, suggesting that uneven repartition of CD107a, in a relatively constant fraction 241 of cells at each division round, does not lead to the emergence of well-defined cellular 242 subsets expressing higher or lower levels of CD107a. We employed the Kolmogorov-243 Smirnov goodness of fit test to determine whether the different curves followed the same 244 distribution or not. The test strongly rejected the hypothesis that the CD107a expression 245 curves follow the same distribution during the first two division rounds (see Materials and 246 Methods section), indicating that during these division events randomly heterogeneous 247 populations were generated. Nevertheless, our test also showed that the Kolmogorov 248 distance decreased when the number of divisions increased, indicating that CD107a-FI 249 density distribution seems to be convergent with a higher number of divisions. To define 250 where variability was located in the curves, we employed the  2 test. The test showed that 251 variability was distributed all over the curves (i.e. for all the CD107a-FI). Together, 252 Kolmogorov-Smirnov goodness of fit and  2 tests revealed a non-stationary variability in 253 the content of CD107a + vesicles in CD8 + T cells during early division events. 254 Taken together, the above results indicate that asymmetric distribution of CD107a + vesicles 255 in telophase is not limited to the first division, but it is rather a stochastic process, inherent 256 to each division, that perpetuates variability in daughter cells.  Figure 6C and Movie 4 (LTR distribution 273 ranged within 40-60% at all time points measured). One CD8 + T cell that distributed in an 274 asymmetry fashion LTR + vesicles is shown in Figure 6D and Movie 5 (LTR distribution 275 ranged above or below 40-60% at all time points measured). Additional examples of cells 276 dividing in symmetric and asymmetric fashion are shown in Figure S5 and Movie 6. 277 Together, the above data support the hypothesis that uneven repartition of lytic granules 278 during division is a stochastic event. 279 Finally, we used a computational approach to establish whether the above-described 280 process might be linked to a random repartition of lytic components into the two nascent 281 daughter cells. We calculated the probability that individual vesicles might fall on the two 282 sides of the division furrow. Using stimulated emission depletion (STED) on CTL stained 283 for GrzB, we estimated that 14 to 65 lytic granules are contained within individual CTL. 284 We next calculated the probability to obtain an asymmetric distribution of lytic granules 285 To assess the consequences of an uneven distribution of lytic compartment on CTL-298 mediated cytotoxicity, we investigated cytotoxic efficacy in CTL expressing high and low 299 lytic granule content. To this end, clonal CTL were loaded with LysoTraker blue, and cells 300 containing high (LysoTracker High ) and low (LysoTracker Low ) levels were FACS sorted. 301 As shown in Figure 7A, sorted LysoTracker High and LysoTracker Low CTL populations 302 maintained their difference in LysoTracker staining at least 24 hours after cell sorting. The 303 cytotoxic efficacy of sorted CTL populations was compared at different effector:target 304 (E:T) ratios by measuring the percentage of killed targets (7-AAD positive targets). For 305 each ratio, LysoTracker High CTL were more efficient than LysoTracker Low CTL in exerting 306 cytotoxicity ( Figure 7B-C), although basal killing (in the absence of peptide stimulation) 307 was comparable between LysoTracker High and LysoTracker Low CTL ( Figure 7C). 308 The above results show that lytic granule content is associated with killing efficacy. 309 Together, they suggest that stochastic uneven distribution of lytic vesicles in dividing CD8 + 310

T cells impact killing behavior. 311
Discussion 312 In the present study we found that, in both freshly isolated peripheral blood CD8 + T cells 313 and clonal CTL, ~ 20 percent of telophasic cells undergoes asymmetric distribution of the 314 lytic compartment into the two daughter cells. Our results establish that CD8 + killing 315 capacity is associated to lytic compartment level and strongly suggest that uneven lytic 316 machinery repartition produces CD8 + T cell populations with heterogeneous killing 317 capacities. Our results demonstrate that the uneven lytic machinery distribution is not related to ACD. 331 In mouse T lymphocytes, ACD has been reported as a mechanism contributing to the 332 small quantity of granules. In other words, pre-packaged molecular components within a 376 few relatively big vesicles might have higher probability to be asymmetrically partitioned 377 in telophase than molecular components dispersed throughout the cytosol. 378 Together, our results point out a mechanism of heterogeneity generation that is purely 379 stochastic and might be a general mechanism for generating heterogeneity in dividing cells.      (Threshold (M02, tubulin, 75), 50-5000, 0-0.5)) based on the α-tubulin 487 staining (to clearly identify the narrow intracellular bridge of highly condensed -tubulin 488 that participates to midbody formation) was employed to distinguish telophases from 489 anaphases or cell-doublets. Finally, the results from both masks were used to manually 490 verify that selected cells were cells unambiguously in telophase.

Live cell imaging 548
For 3D live cell imaging, the T7 GZMB sequence was obtained by PCR amplification as a 549 XhoI-EcoRI fragment from pCMV-SPORT6-GZMB by using XhoI-T7-GZB forward 550 primer and EcoRI-GRZB noSTOP reverse primer (Employed primers: Name: XhoI-T7-551 GzB F 552 28 caaCTCGAGTAATACGACTCACTATAGGGAGACCCGGTACCatgcaaccaatcctgcttctgcc 553 Name: EcoRI-GzB-noSTOP-R caaGAATTCcggcgtggcgtttcatggttttctttatccag). 554 XhoI-EcoRI fragment was cloned as a mCherry-SEpHlurin fusion construct in the 555 pmCherry-SEpHlurin vector to produce the vector pGZMB-mCherry-SEpHluorin 556 available to in vitro T7 transcription. The plasmid pCMV-SPORT6-GZMB and pmCherry-557 SEpHlurin were purchased from Addgene. 558 For efficient transfection of human CTL with tagged molecules allowing to monitor lytic 559 granule repartition during mitosis, we synthetized capped and tailed poly(A) mCherry-560 tagged Granzyme B mRNA by in vitro transcription from the plasmid pGZMB-mCherry-561 SEpHluorin. One microgramme of pGZMB-mCherry-SEpHluorin was first linearized by 562 NotI digestion to be used as templates for in vitro transcription by the T7 RNA polymerase 563 using mMESSAGE mMACHINE T7 Ultra kit as per manufacturer's protocol. 564 Human CTL were transfected using a GenePulser Xcell electroporation system (BioRad). 565 1x10 6 CTL (5days after restimulation therefore in expansion phase) were washed and 566 resuspended in 100l Opti-MEM medium (Gibco) at RT with 2g mCherry-tagged 567 Granzyme B mRNA (square wave electrical pulse at 300V, 2ms, 1 pulse). 18 hours after 568 transfection the transfection efficacy was verified by FACS analysis (typically 50-80%). 569 Transfected CTL were seeded into poly-D-lysine-coated eight-well chambered slides 570 (Ibidi, Munich, Germany) before imaging. Chambered slides were mounted on a heated 571 stage within a temperature-controlled chamber maintained at 37°C and constant CO2 572 concentrations (5%) and inspected by time-lapse laser scanning confocal microscopy 573 (LSM880, Zeiss, Germany with 1 image /30 seconds) for additional 5-6 hours using a Tile   shall observe that we never reject the hypothesis of independence. 635 The Kolmogorov-Smirnov test consists in analyzing if two independent samples follow the 638 same law comparing their cumulative distribution function. We denote the two samples In that case, the probability of an asymmetric division for n granules is then equal to 658 The funders had no role in study design, data collection and analysis, decision to publish, 677 or preparation of the manuscript.