Dynamics of age-related catastrophic mitotic failures and recovery in yeast

Genome instability is a hallmark of aging and contributes to age-related disorders such as progeria, cancer, and Alzheimer’s disease. In particular, nuclear quality control mechanisms and cell cycle checkpoints have generally been studied in young cells and animals where they function optimally, and where genomic instability is low. Here, we use single cell imaging to study the consequences of increased genomic instability during aging, and identify striking age-associated genome missegregation events. During these events the majority of mother cell chromatin, and often both spindle poles, are mistakenly sent to the daughter cell. This breakdown in mitotic fidelity is accompanied by a transient cell cycle arrest that can persist for many hours, as cells engage a retrograde transport mechanism to return chromosomes to the mother cell. The repetitive ribosomal DNA (rDNA) has been previously identified as being highly vulnerable to age-related replication stress and genomic instability, and we present several lines of evidence supporting a model whereby expansion of rDNA during aging results in nucleolar breakdown and competition for limited nucleosomes, thereby increasing risk of catastrophic genome missegregation.


13
Abstract: Genome instability is a hallmark of aging and contributes to age-related disorders such 14 as progeria, cancer, and Alzheimer's disease. In particular, nuclear quality control mechanisms 15 and cell cycle checkpoints have generally been studied in young cells and animals where they 16 function optimally, and where genomic instability is low. Here, we use single cell imaging to 17 study the consequences of increased genomic instability during aging, and identify striking age-18 associated genome missegregation events. During these events the majority of mother cell 19 chromatin, and often both spindle poles, are mistakenly sent to the daughter cell. This breakdown 20 in mitotic fidelity is accompanied by a transient cell cycle arrest that can persist for many hours, 21 as cells engage a retrograde transport mechanism to return chromosomes to the mother cell. The 22 repetitive ribosomal DNA (rDNA) has been previously identified as being highly vulnerable to 23 age-related replication stress and genomic instability, and we present several lines of evidence 24 supporting a model whereby expansion of rDNA during aging results in nucleolar breakdown 25 and competition for limited nucleosomes, thereby increasing risk of catastrophic genome 26 missegregation.

28
Main Text: Each cell cycle involves a delicate choreography of duplicating genetic material and 29 cellular organelles, and apportioning them appropriately between mother and daughter cell. 30 Failures of cell cycle regulation can result in severely compromised fitness or cells that respond 31 improperly to environmental cues and emerge as cancerous precursors (Hanahan and Weinberg, 32 2011). In particular, aneuploidy (the gain or loss of partial or whole chromosomes) can be 33 deleterious to fitness (Beach et al., 2017;Sunshine et al., 2016) and has been implicated in many 34 different types of cancers (Gordon et al., 2012) as well as developmental diseases such as Down 35 Syndrome (Nagaoka et al., 2012). Recent work has also documented extensive damage and 36 genomic rearrangements that can occur from micronuclei or from telomeric crisis (Maciejowski 37 et al., 2015;Zhang et al., 2015), and identified the rDNA sequences as particularly vulnerable to 38 genomic damage (Flach et al., 2014;Xu et al., 2017). Mitotic processes and checkpoints of the 39 budding yeast Saccharomyces cerevisiae have been extensively studied, but the vast majority of 40 work has focused on logarithmically growing young cells (Beach et al., 2017;Dotiwala et al., 41 2007;London and Biggins, 2014;Palmer et al., 1989;Santaguida and Amon, 2015). By studying 42 nuclear dynamics in yeast as they age, we have uncovered a cause of age-associated genomic 43 instability and an active mechanism to maintain nuclear integrity and proper segregation of 44 genetic material in aged cells. 45 We characterized the dynamics of genome replication and partitioning during replicative 46 aging by imaging cells expressing fluorescently tagged histone 2B (Htb2:mCherry) in a 47 microfluidic device over their entire lifespans (Crane et al., 2014). During each cell cycle, the 48 amount of Htb2 in the mother cell nucleus increases by about two-fold and then drops by one-49 half, as the cell enters mitosis and chromosomes are segregated to the newly formed daughter. 50 The vast majority of cell divisions in young cells follow this characteristic pattern ( Figure 1A). 51 As cells age, however, abnormal segregation events become common ( Figure 1B, Videos 1-2, 52 please ensure volume is on for all Videos to hear audio explanation). The single cell trace shown 53 in Figure1C, for example, shows a cell undergoing multiple cell cycles with proper division until 54 an abnormal segregation occurs in which the majority of detectable histones are sent to the 55 daughter cell. These genome-level missegregation (GLM) events result in cell cycle arrest that 56 can range from a few minutes ( Figure 1C-top) to many hours (Figure1C-middle), before they are 57 usually corrected by returning the aberrantly segregated genetic material to the mother cell. The 58 range of arrest durations is broad, with most events resolved within an hour, but some lasting 59 many hours (Figure 1 -figure supplement 1). If corrected by this REtrograde TRansport Nuclear 60 (RETRN) process, mother cells are able to proceed through subsequent divisions, but if not, the 61 mother cells will terminally exit the cell cycle and senesce ( Figure 1C-bottom). 62 To further characterize GLM and RETRN dynamics during aging, cells were imaged 63 over their entire replicative lifespan, with birth events, GLMs, and RETRN events assessed. The 64 probability of a GLM increased dramatically at the end of life ( Figure 1D), with approximately 65 three quarters of mother cells experiencing one or more GLMs ( Figure 1E). About 90% of GLMs 66 were corrected through successful RETRN events, allowing individual mother cells to live 67 approximately 30% longer on average than if all GLMs were terminal (Figure 1 - figure  68 supplement 2). However, even when corrected, mother cells that undergo a GLM are more likely 69 to die in the near future than cells of the same age that have not experienced such an event, and 70 GLMs become increasingly predictive of impending mortality with increasing age (Figure 1 -71 figure supplement 2). To confirm that the histones do indeed co-localize with DNA during these 72 events, we imaged old mothers and observed the dynamics of Htb2 in cells exposed to the DNA 73 stain Hoechst 3342. As can be clearly seen ( Figure 1F, Videos 3-4), both the DNA and histones 74 move in concert during these events. Furthermore, we confirmed that histone levels reflect 75 nuclear DNA abundance in single cells by staining with DAPI and comparing Htb2 levels with 76 DNA content in old cells ( Figure 1 -figure supplement 3). GLM and RETRN dynamics were 77 not influenced by the fluorophore used or which histone is tagged, as the dynamics of both 78 Htb2:mCherry and histone 2A tagged with GFP (Hta2:GFP) did not differ (Figure 1 -figure  79 supplement 4). GLM and RETRN frequency is not an artifact of our imaging protocol, as 80 modifying the excitation power or the cumulative excitation energy exposure had no effect on 81 these observations (Figure 1 -figure supplement 5). For clarity, the strain containing 82 Htb2:mCherry is referred to as wild-type hereafter. 83 In order to observe the nuclear periphery during GLM and RETRN events, we imaged 84 aging cells expressing both Htb2:mCherry and Nup49:GFP, and compared normal divisions 85 (Video 5) with RETRN events ( Figure 1G-top, Video 6) and terminal GLMs ( Figure 1G-bottom, 86 Video 7). The dynamics of the histone missegregation and recovery can be clearly seen in these 87 time-lapse series, and strikingly the mother cells retain an intact nuclear envelope during these 88 events -even if they lose all the chromatin (Figure1G). Passage of the histones fully into the 89 daughter cell is evident from cells co-expressing a bud neck marker (Myo1:GFP) along with 90 Htb2:mCherry (Videos 1-2). Interestingly, during these events both spindle poles often enter the 91 daughter and move in concert with the tagged histones (Figure1H both spindle poles generally remain in the daughter cell ( Figure 1H, Video 9). This can also be 95 observed in videos where tubulin is tagged with GFP (Tub1:GFP), and all of the detectable 96 nuclear microtubules enter the daughter cell during GLMs (Video 10-11).  showing the process of a normal cell division where chromatin (red) doubles during S-phase and is divided between 101 mother and daughter during mitosis. B) Aging cells frequently experience Genome Level Missegregation (GLM) 102 events where most genomic material enters the daughter while the nuclear envelope stays behind. Usually this 103 missegregetion is corrected (top, RETRN event), allowing mother cells to go on to divide and produce more daughters. If not corrected and cytokinesis occurs (bottom), this becomes a terminal event wherein mother cells 105 replicatively senesce. C) Representative single cell traces of mother Htb2 levels showing missegregation (shaded) 106 and active retrograde correction events. Corrections can occur quickly (top), or can take hours to be completed 107 (middle). A GLM becomes terminal (bottom) if it is not corrected. (*) indicates the formation of new buds, and both 108 cells with RETRN events produce additional daughters. AU indicates arbitrary units. D) Missegregation 109 probabilities increase dramatically near the end of replicative lifespan. n=359 mother cells examined. E) Over their 110 entire replicative lifespan, individual mother cells have a greater than 70% chance of having one or more 111 missegregation events. F) Genomic DNA and histones co-localize during GLM events. Two cells expressing 112 Htb2:mCherry and stained with a live DNA dye Hoechst 3342. G) Time-lapse dynamics of a GLM with RETRN 113 correction (top, mother cell replicative age 14) and a terminal missegregation (bottom, mother cell replicative age 114 12) in cells co-expressing Htb2:mCherry and Nup49:GFP. During both GLMs the nuclear envelope is clearly visible 115 in both mother (M) and daughter (D) cells. See Videos 6 and 7. H) Time-lapse dynamics of a GLM with RETRN 116 correction (top, mother cell replicative age 13) and a terminal missegregation (bottom, mother cell replicative age 117 16) in cells expressing Htb2:mCherry and Spc72:GFP. Both spindle poles can be seen to enter the daughter (D) 118 during these events, and during the RETRN event a spindle pole returns to the mother (M). In the terminal 119 missegregation, the spindle pole fails to reenter the mother cell. See Videos 8 and 9. Times are indicated in 120 hours:mins from the start of the displayed time-lapse, not the start of the experiment. Arrows indicate mother cells 121 without visible chromatin.

123
To investigate the temporal and cell cycle dynamics of the genome missegregation and 124 RETRN events, we employed two cell cycle reporters, Whi5:GFP and Myo1:GFP (Di Talia et al., 125 2007). Whi5 prevents exit from G1 when localized in the nucleus, and the transition of Whi5 126 from the nucleus to the cytoplasm signifies the end of G1 ( Figure 2A) (Charvin et al., 2010). 127 Individual young cells proceed in a reliable fashion through the cell cycle, with Whi5 becoming 128 nuclear localized immediately after Htb2 levels fall and the cell enters telophase (FigureS6). By 129 aligning all annotated cell divisions without GLMs ( Figure 2B), the temporal dynamics of 130 histone separation into the daughter cells are clear and immediately followed by the transition of 131 Whi5 from the cytoplasm into the nucleus. In contrast, when a GLM occurs, Htb2 levels fall 132 precipitously in mother cells, but the cell delays the cytoplasmic to nuclear transition of Whi5 133 until after the RETRN correction ( Figure 2C). Because the Whi5 cytoplasm-to-nuclear transition 134 occurs upon activation of the mitotic exit network (Bean et al., 2006;Costanzo et al., 2004), this 135 delay demonstrates that the cell prolongs mitosis until the GLM is corrected.

136
In a complementary fashion, we confirmed that GLMs induced a delay in mitotic exit by 137 following Myo1:GFP localization and abundance. Myo1 is produced at the end of G1 as the cell 138 begins the formation of a new daughter (Bi and Park, 2012;Weiss, 2012). At the end of mitosis, 139 following successful partitioning of the genomic material, Myo1 is degraded during cytokinesis 140 ( Figure 2A, Figure 2 -figure supplement 1). By manually aligning normal cell divisions, the 141 increase of Myo1 levels at the bud neck, and subsequent decrease during cytokinesis, can be 142 clearly seen ( Figure 2D). When the Htb2 levels reach their lowest point, Myo1 levels are at a 143 maximum, but they immediately begin to fall as the cell eliminates the bud neck during 144 cytokinesis. In contrast, in divisions where a GLM occurs, the levels of Myo1 continue to 145 increase even after the Htb2 levels reach their lowest point ( Figure 2E). Only following the 146 RETRN event does Myo1 begin to fall. 147 To determine whether GLMs result from improper spindle attachment and whether 148 RETRN events are caused by activation of the spindle assembly checkpoint, we deleted the gene 149 encoding the spindle assembly checkpoint component Mad3 (mammalian BubR1). This failed to 150 alter the age-related increase in missegregation, and older mad3∆ cells had the same 151 missegregation rates as wild type cells (Figure 2 -figure supplement 2). Taken together, these 152 data differentiate RETRN events and terminal GLMs from prior observations of nuclear 153 oscillations near the bud neck linked to alignment of the spindle poles between mother and 154 daughter (Palmer et al., 1989;Yang et al., 1997;Yeh et al., 2000Yeh et al., , 1995. Similarly, nuclear 155 excursions where a single spindle-pole entered the daughter, but then returned to the mother has 156 been identified in DNA damage checkpoint mutants (Dotiwala et al., 2007); however, when cells 157 arrest as a result of the DNA damage response during metaphase, only the nucleolus enters the 158 daughter while all chromatin is retained by the mother cell (Witkin et al., 2012). A recent report 159 identified segregation of the nucleus and spindle poles into the daughter cell in five aging yeast 160 cells (of the 10 observed), which is likely to be the same phenotype detailed here (Neurohr et al., 161 2018). They also identified elevated rates of chromosome I missegregation in old cells (in ten of 162 the forty cells observed). In prior studies, mutants that missegregate chromosomes or both 163 spindle poles to the daughter cell are unable to correct these events (Finley et al., 2008;Thrower 164 et al., 2003;Yeh et al., 2000Yeh et al., , 1995; however, the RETRN events seen in aged mother cells 165 unambiguously delay mitotic exit and correct these failures by actively transferring chromatin, 166 microtubules, and/or spindle poles from the daughter cell to the mother cell. 167  has been shown to increase nearly twenty fold (Dang et al., 2009). To test whether histone levels 196 reflect ERC abundance, we determined the impact of reducing ERC formation by removing the 197 replication fork block protein Fob1; deletion of FOB1 reduces rDNA recombination and ERC 198 abundance (Defossez et al., 1999;Kobayashi et al., 1998;Sinclair and Guarente, 1997) and the increase in histone levels with age is attenuated in fob1∆ cells ( Figure 3B). Furthermore, 201 although wild-type and fob1∆ mother cells stained with DAPI after 16h of growth had 202 statistically indistinguishable replicative ages ( Figure 3C, median RLS=13, p=0.42), wild-type 203 cells have significantly greater DNA levels ( Figure 3D, p=0.0002). The lower DNA and histone 204 levels in old fob1∆ cells relative to age-matched wild-type is consistent with the model that 205 increasing histone levels are driven by ERC accumulation in aging yeast. If increasing histone 206 levels reflect ERC levels, then histone levels should act as a biomarker that predicts remaining 207 lifespan and could be associated with other age-associated failures. Indeed, within cells of the 208 same replicative age, we determined that histone abundance is negatively correlated with 209 remaining lifespan, and this correlation becomes stronger as cells age (Figure 3 - figure  210 supplement 3). Histone levels are less correlated with mortality in fob1∆ cells, supporting a 211 model where slowing ERC production reduces the underlying pathology linking histone 212 abundance to death.

231
The rDNA is localized to the nucleolus, a phase separated organelle inside the 232 nucleus (Lindström et al., 2018). To examine the relationship between histone dynamics, ERCs, 233 and nucleolar structure, we followed aging mother cells expressing Htb2:mCherry along with a 234 GFP fusion to the nucleolar protein Utp13. In young cells, the nucleolus appears as a nuclear 235 site adjacent to the majority of histones, while in old cells, a large quantity of histones is 236 localized within the nucleolus ( Figure 3C, Videos 13-14). Furthermore, the size of the nucleolus 237 increases dramatically as cells age and often becomes fragmented into multiple foci ( Figure 3D, 238 Videos 13-14), further indicating that the excess histones are localized to ERCs in old cells and 239 confirming earlier work linking ERCs to nucleolar fragmentation (Neurohr et al., 2018;Sinclair 240 et al., 1997;Sinclair and Guarente, 1997 Amon, 2009). By averaging the dynamics of cell cycles that behave normally, it is clear that 254 Cdc14 begins to exit the nucleolus prior to division of genomic material between mother and 255 daughter cells ( Figure 4A, Video 15). In cell cycles that experience GLMs, however, Cdc14 256 remains localized to the nucleolus but is released immediately preceding a RETRN event ( Figure  257 4B, Video 15). This agrees with prior work showing that Cdc14 release during anaphase is 258 required to generate pulling forces within the mother to counteract those in the daughter (Ross 259 and Cohen-Fix, 2004). 260 Cdc14 is specifically required for condensation and segregation of repetitive DNA 261 sequences including the rDNA and telomeres (D'Amours et al., 2004;Sullivan et al., 2004). In 262 order to further explore the consequences of failed Cdc14 release on anaphase dynamics during 263 GLM events, we directly observed Chromosome XII by targeting a LacI:GFP reporter to LacO 264 sites engineered on Chr XII (Ide et al., 2010). During GLMs where the majority of DNA enters 265 the daughter cell, both Chr XII chromatids remain behind in mother cell ( Figure 4C, Video 16-266 17). Furthermore, during these GLMs, the Chr XII sister chromatids appear as a single point, 267 only separating into two distinct foci following a RETRN event ( Figure 4C, Video 16-17). This 268 suggests that delayed Cdc14 activation prevents separation of the rDNA and results in improper 269 condensation and segregation of Chr XII. During this process, as Chr XII remains localized to 270 the nucleolus in the mother cell, the remaining genomic content missegregates to the daughter.

271
We confirmed that this behavior is unique to Chr XII by imaging Chr IV and V during GLMs.

272
Both copies of Chr IV and Chr V are missegregated to the daughter cells with the rest of the 273 chromatin during GLMs ( Figure 4D,E, Videos 18-19). Thus, we propose that age-associated 274 expansion of the rDNA and histone depletion lead to dysregulation of Cdc14 during the 275 metaphase to anaphase transition which results in improper genomic segregation. In both 276 terminal GLMs and RETRN events, Cdc14 eventually exits the nucleolus to trigger anaphase 277 (

280
Dynamics of Cdc14 and Htb2 in mother cells during normal divisions (n=2513). As the cell enters mitosis, Cdc14 281 leaves the nucleolus which triggers anaphase (indicated by the arrow). B) Dynamics of Cdc14 and Htb2 in mother 282 cells during RETRN events (n=298). Although genomic content has left the mother and entered the daughter, Cdc14 283 is still localized to the nucleolus. Shortly following Cdc14 release from the nucleolus (indicated by the arrows), the 284 cells experience a RETRN event. C) Direct observation of Chr XII using LacI:GFP and LacO repeats on Chr XII.

285
When the cell experiences a missegregation event, both chromatids of Chr XII remain behind in the mother until the 286 RETRN event. After this, a single green dot can be seen in both mother and daughter cells. Mother cell replicative

297
During mitosis, cells must remove and then replace nucleosomes on each DNA strand. 298 Because all replicating DNA pulls histones from a common pool, a mismatch between local 299 histone supply and demand could lead to gaps in nucleosome occupancy, which have been 300 previously observed in aged cells (Hu et al., 2014). We hypothesized that expansion of the rDNA 301 due to increasing ERC levels during aging elevates mother cell demand for histones, which is 302 only partially compensated for by the observed increase in histone expression. By this model, 303 histone demand should be lower in old fob1∆ cells compared to age-matched wild-type cells (due 304 to fewer ERCs), which should result in a reduced probability of GLMs. Observations of GLMs 305 in a fob1∆ mutant support this model ( Figure 5A). In particular, fob1∆ and wild-type cells 306 diverge most dramatically following replicative age 10 ( Figure 5A), when the wild-type cells 307 begin to increase histone levels ( Figure 3B). 308 To directly test the mechanistic link between histone competition and GLMs, we 309 genetically manipulated the supply of histones ( Figure  ( Figure 5B). To reduce histone abundance, we deleted SPT21, which encodes a protein that 315 positively regulates histone expression (Dollard et al., 1994;Kurat et al., 2014) and whose loss 316 has been previously shown to reduce histone levels and increase rDNA instability (Eriksson et al., 317 2012;Kobayashi and Sasaki, 2017). In contrast to deletion of either HPC2 or TOM1, deletion of 318 SPT21 caused increased frequency of GLMs in aging mother cells. When averaged over the 319 entire lifespan, spt21Δ cells experienced significantly more GLMs than wild-type cells, while 320 hpc2Δ and tom1Δ cells had significantly fewer events ( Figure 5C). These observations 321 demonstrate that altering histone abundance in aging cells is sufficient to modulate the frequency 322 of GLM events both upward and downward, and support a model whereby rDNA expansion 323 causes increasing competition for a common histone pool in aging cells which drives the 324 dramatic increase in GLMs during aging ( Figure 5D). 325

339
Because function declines in many different and subtle ways during aging, catastrophic 340 failures and homeostatic systems like the those uncovered here may only be detected in aged 341 organisms. Imaging of individual yeast cells through microfluidic trapping allowed us to observe 342 GLMs that occur in most mother cells at some point during their lives.

544
Microfluidics 545 Cells were imaged using a PDMS microfluidic flow chamber modified from an earlier design 546 (Crane et al., 2014) to increase retention over the replicative lifespan of the mother cells. The 547 microfluidic device was composed of multiple chambers in the same fashion as (Granados et al.,548 2017), which allowed individual genotypes to be exposed to identical environments and imaged 549 in the same experiment while being physically isolated. Cells were loaded according to 550 previously published methods (Granados et al., 2017). A volumetric flow rate of 3-7 µL/min per 551 chamber was used, with the flow rate starting low, and increasing during the experiment to 552 improve mother cell retention and to ensure that cells do not aggregate, which can clog the 553 device.

555
Microscopy 556 Cells were imaged using a Nikon Ti-2000 microscope with a 40X oil immersion objective with a 557 1.3 NA and using the Nikon Perfect Focus System. An enclosed incubation chamber was used to 558 maintain a stable 30C environment for the duration of the experiment. Two Aladdin syringe 559 pumps were used for media flow. An LED illumination system (Excelitas 110-LED) was used to 560 provide consistent excitation energies, and to minimize the exposure, illumination was triggered 561 by the camera. Images were acquired using a Hammamatsu Orca Flash 4.0 V2. The microscope 562 was controlled by custom software written in Matlab® and Micromanager. 563 564 Images were corrected for illumination artifacts in two stages. First, to correct for individual 565 differences in the pixel biases, 1,000 images were acquired with no illumination, and the 566 individual pixel means were determined. Second, to correct for flatness of field, a fluorescent 567 dye was added to a microfluidic device instead of using a slide with dye. Using a slide containing 568 dye introduces a large amount of out-of-focus light, which results in an underestimation of the 569 field curvature. In order to compensate for the microfluidic features, 1,000 images were acquired 570 each with a small offset in the x and y positions. Images were then dilated, and the median value 571 at each location was used. Thus, for each image, the camera bias for that pixel was subtracted, 572 and then it was multiplied by a flatness of field correction factor. 573 574 Images were acquired at 5 min intervals for bright-field and fluorescent channels. The 575 fluorescence excitation power was set to 25% for all imaging except the GFP tagged histones, 576 where it was set to 12%. Fluorescence and brightfield light was activated during image 577 acquisition and all other lights in the room were turned off. For bright-field, 3 z-sections were 578 acquired with 2.5 µm intervals, exposure times of 30 ms and were used for automated 579 segmentation and tracking. For the fluorescent channels, 3 z-sections were acquired with 1.5 µm 580 spacing. GFP images were acquired using a Chroma ET49002 filter set, and mCherry images 581 were acquired using a Chroma ET49306. GFP images were acquired using exposure times of 582 60ms for all proteins except Htb2 and Hta2 which were acquire using a 30ms exposure time. 583 mCherry images were acquired using a 60 ms exposure time. These imaging conditions were 584 found to work as a reasonable compromise between the desire for frequent, dense imaging to 585 enable identification of missegregations and retrograde transport, while also minimizing 586 phototoxicity. We performed control experiments to verify that these exposure conditions did not 587 affect the rates of genomic missegregation or replicative lifespan (Figure 1 -figure supplement 588 5). Each strain was imaged in multiple independent experimental runs, each with approximately 589 equal numbers of cells.

591
Data processing and single cell scoring 592 Following data acquisition, cells were identified and tracked using previously published 593 software (Bakker et al., 2017). This identified the cell outline, and performed initial tracking of 594 the cells through time. To ensure that only young, healthy cells were assessed, we only used cells 595 that were identified in the first three hours of the experiment. Birth events for these cells were 596 then manually scored, and any errors in tracking were corrected. This was all done using the 597 bright-field images. Birth events were scored by multiple observers who were blinded. Because 598 individual cells can be lost from traps prior to death, it can be challenging to know whether using 599 censored (lost) cells is most appropriate. In the supplementary information, all data is presented 600 with and without censoring. In the main text, plots aligned based on increasing age used all cells 601 present at that age, even if they were later lost from the device. For plots aligned by death, only 602 cells that had either died or senesced (failed to initiate a new cell division but did not visibly lose 603 cell wall integrity during the experiment) were used. Because censoring in lifespan experiments 604 relies on the assumption that losses are unbiased, we provide replicative lifespan curves both 605 including and excluding censored cells for all strains. Censoring does not change the 606 interpretation or statistical outcome of any of the experiments presented here. 607 608 For each cell division, the mean histone levels over that cell cycle were used. For cell cycles that 609 last more than three hours, only the first three hours were used to determine the histone levels.

610
This ensured that cells which have a terminal missegregation at end of life do not show an 611 inaccurately low histone level for the last cell cycle.

613
Following manual scoring of birth events, the fluorescent channel containing the histones was 614 used to observe the missegregation dynamics. To ensure consistent scoring across experiments 615 and eliminate bias, information about the experiments was masked from the scorer until after the 616 data was evaluated. Following genome level missegregations, RETRN events were defined as 617 where the histone fluorescence decreased in the daughter cell while simultaneously increasing in 618 the mother cell. This prevented any changes in focus or gradual fluorescence increases due to 619 fluorophore maturation from being inadvertently scored as a RETRN event. RETRN events were 620 scored at the timepoint that the histones return to the mother cell. During cell cycles where cells 621 had multiple retrograde events during the same cell cycle, only the final RETRN event was 622 scored. Events were scored as terminal missegregation events if, prior to a correction, the 623 daughter cell visibly separated from the mother cell (indicating cytokinesis) or if the mother 624 died.

626
Fluorescence quantification 627 Quantification of the level of protein localized to either the nucleolus (Cdc14) or the nucleus 628 (Whi5), was done using a measure of how asymmetrically distributed the fluorescent signal was. 629 Specifically, we used average brightness of the top 2% of pixels, divided by the cell median. By  630 normalizing to median fluorescence, we corrected for any changes in fluorescence that could 631 occur as a result of photobleaching. This method has been used previously as an accurate 632 measure of the fraction of protein that is nuclear localized (Cai et al., 2008;Granados et al., 633 2017). For Myo1 quantification, we used the mean fluorescence level along the periphery of the 634 cell as segmented. The nucleolar localized fluorescent protein (Utp13:GFP) was used to segment 635 the nucleolus and determine the nucleolar size. This was done by calculating a segmentation 636 threshold using Otsu's threshold (Otsu, 1979) for each cell applied to the maximum projection of 637 the fluorescent image stack. This approach was found to have good agreement with earlier 638 estimates of nucleolar size in young cells grown in glucose (Jorgensen et al., 2007).

640
Because we are utilizing an epi-fluorescent microscope and as a result of out-of-focus light, 641 mean fluorescence is not directly linked to protein concentration (Bakker, 2016). Because the 642 fluorescence levels change dramatically during aging, segmenting out the nucleus to determine 643 fluorescence levels could introduce significant biases as the cells age. Additionally, taking the 644 mean fluorescence level could introduce significant errors depending on the segmentation 645 accuracy. Instead, to normalize for changes, we sum a constant area within each cell. This 646 prevents segmentation (of either the nucleus or cell) from contributing to the determination of 647 histone amount. Pixels are sorted from high to low, and the top fraction corresponding to a circle 648 with diameter of 3.8 µm are used to calculate the mean.

650
Yeast Strains and Growth 651 The GFP strains were all acquired from the yeast GFP collection (Huh et al., 2003). The 652 Htb2 Htb2:mCherry. Complete list of strains available in Table S1.

661
Prior to each experiment, single colonies were picked into SC media (Sunrise Biosciences) with 662 2% dextrose. Cells were grown overnight, and then diluted 1:200 in fresh media and grown for 663 5-6h. Prior to loading into the microfluidic device, 0.5 mL of SC 2% dextrose with 0.5% BSA 664 was added to each 5 mL culture to prevent the cells from adhering to the PDMS during loading.

665
During experiments, SC media with 2% dextrose and 0.1% BSA was used, and cells were 666 imaged for 72h.

668
Statistical Analysis 669 Error bars in the figures which contained bar plots were generated by bootstrapping with 670 replacement, and then determining the 95% confidence intervals. Error bars in figures with line 671 plots are standard error. Statistical significance for lifespan was determined using the log-rank 672 test. Log-rank test was performed with, and without, censored cells that were lost prior to 673 senescence or death. To compare distributions (such as numbers of missegregation events over 674 the lifespan), a two-tailed t-test assuming equal variance was used. Correlations between histones 675 or missegregation events and remaining replicative lifespan were calculated with the Spearman 676 correlation using the population of cells alive at each replicative age. Correlation between 677 Htb2:mCherry levels and DAPI staining were also done using the Spearman correlation. 678 679

680
Differences between censored and uncensored survival data 681 Frequently in experiments or clinical studies that involve the generation of survival curves, some 682 samples will be removed from the population under observation. For example, a patient may 683 leave a study not because of death, but because they move to a different country. This can be 684 treated in a relatively straightforward manner statistically including these individuals in the 685 analysis until the point that they are lost (or censored). This relies on the assumption that there is 686 no bias in whether a sample is lost or retained. A recurring concern with microfluidic aging 687 experiments involving yeast is whether there is a bias in how cells are lost or retained. This 688 appears especially important when the mutation or transgene affects cell morphology or cell 689 cycle, as this can result in a bias in which cells are lost from the traps. To reduce the likelihood 690 that our observations were directly affected by loss rates, in the main text we have plotted all 691 cells that were present at that replicative age for plots from birth. Thus, if a cell was lost at 692 replicative age 20, it was included in the plots until age 19. In the supplementary materials we 693 have also repeated every plot, but only using cells that died in the microfluidic device. Given that 694 this is an altered population distribution and smaller number of cells, these plots are slightly 695 different, but they do not affect the conclusions. Furthermore, in the plots where cells are aligned 696 by birth rather than death, we are forced to only use cells that die in the device. For replicative 697 lifespans shown in the supplementary, we include survival curves with and without censored 698 cells.

700
Aligning cells from birth or from death 701 Cells can be aligned either by birth (counting up from replicative age = 0), or by death (counting 702 back from death). Either processing makes some assumptions about how similar cells are to one 703 another. If cells are most similar to each other when they are born, aligning by birth makes sense, 704 and as the replicative age increases, the number of samples decreases because cells are removed 705 by death or senescence. In contrast, assuming that cells are similar at death implies that the 706 phenotype of interest is most similar as cells approach death. For example, the average time cells 707 take to proceed through each cell cycle increases geometrically when cells are aligned by birth, 708 but exponentially when aligned by death. In the supplementary figures, we show the population 709 means aligned using both approaches. 710 711 Differences between Htb2:mCherry and Hta2:GFP 712 Although both of these strains were found to have similar numbers of missegregation events 713 during their lifetimes, and similar fractions of these events were corrected (Figure 1 -figure  714 supplement 4B,C), there are subtle differences between the strains. Most notably, the strain with 715 Hta2:GFP had what we consider to be a normal lifespan for this background (median RLS= 21 716 for cells that died/senesced in the device, median RLS=24 including censored cells, Figure 1 -717 figure supplement 4D,E). The strain with Htb2:mCherry, however, had a somewhat shorter 718 lifespan (median RLS = 16 for cells that died/senesced, median RLS=18 including censored 719 cells, Figure 1 -figure supplement 4D,E). Removing FOB1, however, results in an increase of 720 the replicative lifespan of this strain by ~30% (Figure 3 -figure supplement 3), which is in line 721 with results from literature (McCormick et al., 2015). Furthermore, the increase in replicative 722 lifespan as a result of increased histone transcription has been less thoroughly studied, but our 723 results are in line with those previously reported by another group (Feser et al., 2010;Kruegel et 724 al., 2011). Thus, although there is an unexpected reduction in lifespan for the Htb2:mCherry 725 strain, we do not believe that it affects our results.

727
Likewise, as shown in the main text, we determined the correlation between missegregation 728 events and remaining lifespan at the single cell level. The correlation is between the binary 729 presence or absence of a missegregation event during a cell cycle and the remaining lifespan. 730 Strikingly, as shown in Figure 1 -figure supplement 2, for both strains, the correlation between 731 missegregation events and remaining replicative lifespan is the same for both Htb2:mCherry and 732 Hta2:GFP. This is in spite of the difference in absolute lifespan between the two strains.

734
Because GFP fluorescence is much more affected than mCherry fluorescence by changing pH 735 ( Shaner et al., 2005), and the pH of the cytoplasm in aging yeast has previously been shown to 736 increase (Henderson et al., 2014), we chose to perform the majority of the experiments using 737 mCherry. This ensured that any changes in pH homeostasis during aging would not affect our 738 measurements of histone levels.

740
GFP Tagged Histones and Correlation with remaining RLS 741 As discussed in the main text, we imaged GFP tagged histones acquired from the GFP collection 742 (Hta2, Htb2). This allowed us to determine that the increase in histone levels was a general 743 aspect of aging physiology, not confined to specific histones. In order to determine whether 744 histone levels were accurate biomarkers that predicted remaining lifespan, we used the single cell 745 measurements collected for both strains (Figure 3 -figure supplement 1). At each replicative 746 age, cells still alive are used to determine the correlation between their remaining replicative 747 lifespan and the single cell histone levels. Thus, for prediction of remaining lifespan at 748 replicative age 5, only cells that bud more than 5 times are included. The negative correlation 749 between histone levels and remaining lifespan is similarly true for the GFP tagged histones 750 imaged: Hta2, Htb2 (Figure 3 -figure supplement 1A). Furthermore, all histones are similarly 751 predictive of remaining lifespan, and become increasingly predictive as cells age.

753
To provide a more complete picture of the data acquired, we plotted the GFP mean histone levels 754 where single cells are aligned based on their replicative ages (Figure 3 -figure supplement 1B). 755 As the population ages, the underlying distribution changes as cells die and are removed from the 756 pool. This creates some variability with the increasing replicative age of the population. When 757 individual cells are aligned by death, the population mean for histone levels at replicative ages 758 preceding death looks quite different (Figure 3 -figure supplement 1C). By aligning cells by 759 death, it is clear that histone levels begin to increase around 5-10 replications prior to death.

760
When including all cells, including those that were lost from the traps, cells expressing GFP 761 tagged histones have similar replicative lifespans (Figure 3 -figure supplement 1D,E).

763
Single Cell DNA Levels and FOB1 764 As discussed in the main text, we determined the correlation between histone levels and DNA by 765 comparing, in individual cells, levels of Htb2:mCherry with levels of DAPI. To do this, mother 766 cells were aged in the device for 16h, and then fixed with ethanol and stained with DAPI. To 767 determine whether the accumulation of ERCs with age leads to a detectable increase in DNA 768 content, we compared DAPI staining of wild-type and fob1∆. We acquired age-matched mother 769 cells by using a multi-chamber device to simultaneously age wild-type and fob1∆ cells. 770 Comparing mothers that had grown in the device for 16h (median of 13 divisions and thus 771 middle aged), we confirmed that DNA levels increase in wild-type cells are significantly higher 772 compared with fob1∆. The confirmed that the deletion of FOB1 decreased DNA content in aging 773 cells, and not just histone levels. By manually scoring the number of divisions each strain had up 774 until the point of ethanol fixation, we determined that there was no difference between strains in 775 the number of daughters ( Figure 3C, p=0.42). Although the replicative ages for each strain were 776 statistically indistinguishable, the wild-type strain had a statistically significant increase in the 777 DAPI staining levels ( Figure 3D, p=0.0002). Thus, fob1∆ cells accumulate both histones and 778 excess DNA at lower rates compared with wild-type cells. Furthermore, we compared Htb2 and 779 DAPI levels at the single cell level, and histone levels are highly predictive of DNA content at 780 the single cell level (Figure 1 -figure supplement 3).

782
Manipulating histone demand (fob1∆) 783 To provide a more complete picture of how Htb2 levels changed during aging in fob1∆ cells 784 compared with wild-type, we performed the same procedure as with the GFP strains but using 785 only the cells that die in the device to complement the plots in the main text. These plots look 786 very similar to the plots using censored cells indicating that censoring doesn't alter the 787 population distribution. By aligning cells based on the replicative age, and plotting the mean 788 Htb2:mCherry levels of the population, it is clear that there is a dramatic difference between 789 wild-type and fob1∆ cells in Htb2 levels ( supplement 3G).

802
Manipulating histone supply (tom1∆, hpc2∆ and spt21∆) 803 To provide a more complete picture of how the manipulation of the histone supply affected 804 lifespan and Htb2 levels, we performed a similar procedure as previously described. Spt21 805 positively regulates the transcription of all histone genes, and Hpc2 negatively regulates histone 806 genes. The protein Tom1 is involved in the ubiquitination and degradation of excess histone 807 proteins. Thus by deleting HPC2, we upregulate histone transcription, deleting SPT21 suppresses 808 histone transcription and deleting TOM1 allows excess histone proteins to remain in the cell. 809 Similar to the figures in the main text, we used only cells that die in the device to plot  shown), and become more predictive (more strongly anti-correlated, as shown) with increasing 832 replicative age.  across the entire range of Htb2:mCherry fluorescence, as even cells that have the highest 853 Htb2:mCherry levels have correspondingly high DAPI staining. N=800 individual mother cells. 854 increase in the probability of GLM events in the 5-6 divisions preceding death. Error bars are 858 standard error. B) There is no difference between Htb2:mCherry and Hta2:GFP with respect to 859 the fraction of cells that experience GLM events. Error bars are 95% confidence intervals from 860 bootstrapping over individual cells. C) The distribution of the number of GLM events that cells 861 experience over their lifetime is similar between the strains but statistically different (p=0.02 862 two-tailed t-test), with Htb2:mCherry cells experiencing, on average, more events over their 863 lifetime. D) Replicative lifespan curves for Htb2:mCherry including censored cells (pink) and 864 excluding (red). E) Replicative lifespan curves for Hta2:GFP including censored cells (light 865 blue) and excluding (blue). 866 determine whether cells were affected by the cumulative exposure to fluorescence excitation 869 energy, we compared GLM rates in cells imaged over their entire lifespans (blue) with those only 870 imaged after they were already aged for 20 hours (red), corresponding to a median replicative 871 age of 16 generations. This is equivalent to approximately 75% of the median lifespan of this 872 strain. To compare these cells to the control, we only quantified GLMs that occurred after 20h. 873 Error bars are 95% confidence intervals generated by bootstrapping with replacement over all 874 cells. No difference was detected in cells imaged continuously over their entire lifespan or only 875 after 20 hours, indicating that there is no cumulative effect of the exposure to fluorescence 876 excitation light on the frequency of GLMs or the ability of cells to correct these through RETRN 877 events. 878 879 showing the temporal dynamics of the proteins used to characterize missegregation events. Whi5 882 exits the nucleus to initiate START and move the cell into S phase. It re-enters the nucleus at the 883 end of mitosis. B) A representative trace of a single cell expressing Htb2:mCherry and 884 Whi5:GFP. Histone levels (pink) increase, and then fall during mitosis. Arrows indicate the 885 timepoint before Whi5 transitions from the cytoplasm to the nucleus. (*) indicate raw Htb2 886 measurements, which were smoothed using a moving average for legibility (solid line). C) 887 Schematic showing the temporal dynamics of Myo1. Myo1 is produced at the end of G1 and 888 localizes to the bud neck until cytokinesis ends mitosis. D) Representative single cell trace of a 889 cell containing Htb2:mCherry and Myo1:GFP. (*) indicate raw Htb2 or Myo1 measurements, 890 which were smoothed using a moving average for legibility (solid line). Arrows indicate the 891 lowest point of Htb2 levels following mitosis. 892 the device. C) When only cells that die in the device are aligned by death, both mad3∆ and wild-898 type cells experience a similar age-related increase in GLM rates that begins around 5 divisions 899 prior to death. D) At the single cell level, GLM events are equally strongly correlated with 900 impending mortality in both mad3∆ and wild-type cells. Dots show correlation between a GLM 901 and the remaining lifespan at that replicative age for that genotype. To show trends, these have 902 been smoothed with a moving average for legibility (solid line). E) RLS curve including 903 censored cells (used in A). F) RLS curve excluding censored cells (used in B-D). All error bars 904 are standard error. 905  cells aligned by birth and excluding cells that were censored (See Figure 3B) B) Mean 921 missegregation rates of wild-type and fob1∆ mother cells aligned by birth and excluding cells 922 that were censored (See Figure 4A). C) At the single cell level, missegregation events are less 923 strongly anti-correlated with remaining lifespan in fob1∆ compared with wild-type. The 924 correlation curve for fob1∆ cells appears parallel to wild-type, but with an offset that is roughly 925 proportional to the increase in replicative lifespan we observed for the fob1∆ cells. Excludes 926 censored cells. Dots show correlation between an individual histone and the remaining lifespan 927 at that replicative age. To show trends, these have been smoothed with a moving average (solid 928 line). D) Mean Htb2 levels of mother cells aligned by death instead of birth, excluding censored 929 cells. E) Probability of missegregation events for wild-type and fob1∆ cells, but aligned by death 930 instead of birth and excluding censored cells. F) Survival curve of wild-type and fob1∆ cells 931 containing only cells which are not lost and censored. These are the cells used in A-E. Survival 932 curves comparing wild-type cells and cells lacking FOB1. G) Survival curve including censored 933 cells lost during the experiment. Compared with wild-type cells, fob1∆ mutants live longer 934 (p<1E-10, log-rank test). These are the cells used in the main figures. N-values report the number 935 of cells, and error bars for C,D,F are standard error. 936