Lipid droplet dynamics during the budding yeast cell cycle influence the timing of cell cycle START

8 Recent work has revealed that metabolism is dynamic over the budding yeast cell cycle , showing that 9 the NAD(P)H autofluorescence oscillates , and protein and lipid biosynthesis are dynamic. Dynamic 10 storage and liquidation of neutral lipids over the cell cycle could contribute to the se metabolic 11 dynamics. However, the dynamics of neutral lipids over the cell cycle are as yet unknown. To elucidate 12 them, we established mNeonGreen-Tgl3 and Pln1-mNeonGreen as protein markers for lipid droplets 13 (LDs) and determined LD dynamics during the cell cycle with single-cell time-lapse microscopy . We 14 found oscillations in the LD number over the cell cycle, with a notable trough around START. Deletion of 15 the genes responsible for either the synthesis of triacylglycerol (TAG) or its mobilisation from LDs 16 lowered LD numbers and abolished the oscillation in LD number. Moreover, in these deletion mutants, 17 we found START to be delayed , suggesting that the mobilisation of TAG from LDs is required for its 18 timely occurrence. The influence of LD dynamics on the timing of START emphasises that research 19 studying cell cycle commitment should consider storage lipid metabolism as a potential contributor to 20 cell cycle START. 21 22

23 Proper control of cell growth and division is fundamental for cells. In eukaryotes, the cyclin/cyclin-24 dependent kinase (CDK) machinery is considered the primary regulator of this process (Mendenhall & 25 Hodge, 1998). However, recent work in S. cerevisiae suggests that an intrinsically dynamic metabolism 26 could also play a role in cell cycle regulation. For instance, CDK-independent oscillations in NAD(P)H 27 levels were found essential for cell division (Papagiannakis et al., 2017) a nd an increase in protein 28 biosynthesis during the G1 phase was found to drive commitment to the cell cycle by temporarily 29 increasing Cln3 levels (Litsios et al., 2019) . As the dynamics of metabolism evidently are also involved in 30 cell cycle control, it is important to precisely understand the metabolic dynamics and the ir underlying 31 cause in order to investigate the precise mechanisms through which metabolism controls cell cycle 32 progression. 33 Multiple metabolic processes oscillate along the cell cycle: the glycolytic flux peaks between cytokinesis 34 and late G1 phase (Monteiro et al., 2019) and mobilisation of stored carbohydrates occurs during G1 35 (Müller et al., 2003; Silljé et al., 1997; Zhao et al., 2016). Lipid biosynthesis peaks during S/G2/M 36 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 2 (Takhaveev et al., 2023), coinciding with increased protein levels of the biosynthetic enzymes for both 37 ergosterol (Blank et al., 2020) and fatty acids (Blank et al., 2017) and increased triacylglycerol levels 38 (Blank et al., 2020) in this phase. 39 Besides lipid synthesis, lipid storage could also be dynamic during the cell cycle. Neutral lipids, namely 40 triacylglycerols (TAG) and steryl esters (SE), are stored in lipid droplets (LDs), which are critical during 41 growth transitions. When budding yeast enters stationary phase, TAGs are synthesised and stored in LDs 42 (Czabany et al., 2008; Markgraf et al., 2014). In contrast, when yeast cells exit stationary phase, these 43 neutral lipids are mobilised and LD size and number decrease (Ganesan et al., 2020; Markgraf et al., 44 2014; Ouahoud et al., 2018). Furthermore, mutants deficient in TAG mobilisation resume growth more 45 slowly than the wild type when cells grown to stationary phase are shifted into fresh medium (Kurat et 46 al., 2009; Ouahoud et al., 2018) . Based on these findings, we wondered whether neutral lipid storage 47 could also be dynamic during the cell cycle and if so, whether these dynamics would affect cell cycle 48 progression. 49 In this study, we established protein markers for LDs and used th em to uncover cell cycle oscillations of 50 the number of LDs. These oscillations show a trough around cell cycle START and a peak midway through 51 S/G2/M, indicating that LDs are first depleted and then re-formed as the cell cycle progresses. In strains 52 lacking enzymes for TAG synthesis or mobilisation, the dynamics of the LD number are lost and START is 53 delayed. Our findings indicate that mobilisation of TAG from LDs contributes to the timely occurrence of 54 START, suggesting that LD dynamics play a role in cell cycle control. 55

56 Establishing protein markers for dynamic, dye-free identification of lipid droplets 57 To elucidate the dynamic behaviour of lipid droplets during the cell cycle in S. cerevisiae , we used 58 fluorescence microscopy in combination with a microfluidic setup (Huberts et al., 2013; Lee et al., 2012), 59 which allows the generation of dynamic single-cell data over several cell cycles under constant growth 60 conditions. To visualise LDs, we did not use chemical stains, since this would have required the 61 continuous addition of a dye to the growth medium to counteract the dilution of the dye as cells grow. 62 Instead, to achieve dye-free identification of LDs, we chose to tag proteins of the LD proteome (Grillitsch 63 et al., 2011) that localise to LDs and no other organelles (SGD, 2017), with mNeonGreen and use these 64 tagged proteins as reporters for LDs. Using image databases of S. cerevisiae with GFP-tagged proteins 65 (Huh et al., 2003; Weill et al., 2018), we selected proteins of the LD proteome whose LD localisation 66 remained unchanged after introduction of a fluorescent tag. Specifically, we selected Pln1, Tgl3 and Rrt8 67 as candidate protein markers for LDs. Pln1 stimulates the accumulation of TAG and stabilises LDs (Gao et 68 al., 2017). Tgl3 is a TAG lipase and thus mobilis es neutral lipids from LDs (Athenstaedt & Daum, 2003; 69 Chauhan et al., 2015) and Rrt8 has been implicated in spore wall assembly (Lin et al., 2013) and 70 transport of plasma membrane proteins (Ueno et al., 2016). 71 To verify that these candidate reporter proteins indeed localise to LDs and can thus serve as LD markers , 72 we fixed cells with formaldehyde, stained LDs with the red fluorescent dye BODIPY- TR and then 73 assessed colocalisation between the stained LDs and the mNeonGreen-tagged marker proteins. Given 74 that the chosen marker proteins have different LD-related functions, we considered that certain marker 75 proteins might localise only to a subset of all LDs, or reversely, could have a localised function beyond 76 LDs. 77 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 3 Before testing the colocalisation between LDs detected with BODIPY-TR and the marker proteins , we 78 first ensured optimal alignment between images recorded in the green and red fluorescence channels 79 using a bilinear transformation approach. As LDs have an average diameter of 0.4 µm (Czabany et al., 80 2008), corresponding to less than three pixels in our microscopy setup , even minor misalignment 81 between the two fluorescence channels would have significantly distorted the colocalisation analysis. 82 We then used wild-type cells stained with BODIPY-TR to show that the red BODIPY-TR signal is not 83 detected in the GFP channel (Figure S1A-D) . Furthermore, we compared the fluorescence intensity 84 measured in the GFP channel between a wild-type control, i.e. autofluorescence, and the three LD 85 reporter strains and verified that bright spots of autofluorescence occasionally present in GFP images of 86 wild-type cells are not detectable in the LD reporter strains with the applied spot detection algorithm 87 (Terpstra et al., 2024) and detection settings (Figure S1E-G) . The following colocalisation analysis 88 therefore is not confounded by the detection of GFP puncta that are due to autofluorescence instead of 89 an LD marker protein, or artefacts of BODIPY-TR fluorescence detected in the GFP channel. 90 Next, to perform the actual colocalisation analysis, we imaged cells stained with BODIPY-TR that also 91 expressed one of the LD reporter fusion proteins Pln1-mNG, mNG-Tgl3 and Rrt8-mNG. We detected 92 puncta of the mNeonGreen-tagged LD marker proteins in the GFP channel and LDs stained with BODIPY-93 TR in the RFP channel and saw that puncta of the reporter proteins and puncta identified with BODIPY-94 TR were often found at similar locations within the cell (Figure 1A). To quantify colocalisation between 95 puncta of BODIPY-TR and puncta of the LD marker protein candidates, we differentiat ed between 96 ‘certain’ and ‘ambiguous’ colocalisation. Colocalisation was classified as ‘certain’ if puncta in the tw o 97 imaging channels had identical midpoints or were located at most one pixel apart; ‘ambiguous’ 98 colocalisation refers to puncta whose midpoints were at most two pixels apart. In the following, we 99 consider both the ‘certain’ and ‘ambiguous’ categories as colocalised. 100 We first assessed the fraction of BODIPY-TR puncta that colocalise with a punctum of an LD reporter 101 protein. We found that 82%, 57% and 41% of BODIPY-TR puncta have a corresponding punctum of Pln1-102 mNG, Rrt8-mNG and mNG-Tgl3 , respectively (Figure 1B) . We also determined the fraction of reporter 103 protein puncta that colocalise with a BODIPY-TR punctum and found that 60%, 59% and 54% of protein 104 puncta colocalise with a punctum of a stained LD for the marker proteins Rrt8-mNG, Pln1-mNG and 105 mNG-Tgl3 (Figure 1B). These data show that not all detected LDs stained with BODIPY-TR are also visible 106 as a marker protein punctum , consistent with the existence of subpopulations of LDs with distinct 107 proteomes (Eisenberg-Bord et al., 2018) and lipid content (Meyers et al., 2016). Furthermore, the data 108 show that not all puncta of t he LD reporter proteins colocalise with an LD stained with BODIPY-TR . 109 Protein puncta of Pln1 without a corresponding BODIPY-TR puncta could be LDs that are still too small 110 for detection with BODIPY-TR , consistent with the finding that appearance of Pln1 puncta often 111 precedes BODIPY-TR puncta by a few minutes (Gao et al., 2017). The Rrt8-mNG and mNG-Tgl3 puncta 112 without a corresponding BODIPY-TR punctum might be indicative of alternative functions of these 113 proteins, outside LDs. 114 Overall, our results show that LDs are heterogeneous, with different marker proteins likely representing 115 distinct subsets of LDs. It is conceivable that one reporter could miss a subset of LDs that another 116 reporter could identify . As all three tested LD markers are endogenous proteins with their own 117 biological functions, it is also possible that their dynamics in part reflect other biological functions, 118 besides reporting the dynamics of LDs. Finally, as the localisation of Pln1-mNG is most comparable to 119 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 4 that of the LD dye BODIPY-TR, Pln1-mNG is the most suitable LD marker to investigate whether there are 120 LD dynamics during the cell cycle. 121 122 123 Figure 1. Establishing protein markers for dynamic, dye-free identification of lipid droplets. (A) Cells expressing a candidate LD 124 reporter protein tagged with mNeonGreen (Pln1-mNG, mNG-Tgl3 or Rrt8-mNG) were fixed with formaldehyde, stained with 125 BODIPY-TR to visualise LDs and then imaged (upper row). Puncta of marker proteins and LDs stained with BODIPY-TR were 126 detected with a spot detection algorithm (lower row); (B) Colocalisation between LDs stained with BODIPY-TR and puncta of the 127 three candidate LD reporter proteins Pln1-mNG, mNG-Tgl3 and Rrt8- mNG. The left circle of each Venn diagram represents the 128 marker protein puncta identified in the GFP channel image and the right circle represents puncta of BODIPY- TR in the RFP 129 channel image. The orange overlapping region between the two circles represents the colocalised puncta. The white regions 130 represent puncta whose colocalisation is ambiguous: while puncta in the two fluorescence channels have midpoints close 131 together, their colocalisation is likely, but not certain. The following number of cells and puncta were assessed in the 132 colocalisation analysis: Pln1-mNG: 153 cells, 304 BODIPY-TR puncta, 424 reporter protein puncta; mNG-Tgl3: 217 cells, 345 133 BODIPY-TR puncta, 263 reporter protein puncta; Rrt8-mNG: 245 cells, 323 BODIPY-TR puncta, 302 reporter protein puncta. 134 135 Number of lipid droplets oscillates over the cell cycle 136 Next, we performed time-lapse microscopy experiments to determine the number and size of LDs over 137 the cell cycle using Pln1-mNG and mNG-Tgl3 as markers for LDs. Notably, we did not employ Rrt8-mNG 138 as an LD reporter in our time-lapse experiments. Due to the low er expression of Rrt8 compared to Pln1 139 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 5 and Tgl3 (Breker et al., 2013; Chong et al., 2015; SGD, 2017), imaging required higher light exposure, 140 resulting in prolonged cell cycle durations (Figure S2) , likely due to phototoxicity. By studying LD 141 dynamics with two distinct protein reporters, we aimed to elucidate the more general cell cycle 142 dynamics of LDs, independent of specific functions of the reporter proteins . Therefore, we focused on 143 global similarities in any eventual dynamics observed with the two markers, disregard ing small 144 differences between the two reporters. 145 We used an automated spot detector algorithm (Terpstra et al., 2024) to detect LDs and to estimate 146 their size in the time-lapse images. We normalis ed the number of LDs detected within a cell to the area 147 of its cross-section. To project the detected LD characteristics on a common cell cycle progression 148 coordinate, we aligned all LD cell cycle trajectories for mitotic exit, START and budding. Finally, we 149 applied Gaussian process regression to predict the average cell cycle dynamics of the LD number 150 normalised to the cell cross-area, and of the LD size. 151 With both LD reporter proteins , we found that t he area-normalised LD number oscillates over the cell 152 cycle with a minimum around START and a maximum between budding and mitotic exit (Figure 2A-B). 153 Density plots showing all data points reveal the same cell cycle dynamics (Figure S3A-B) as the 154 population average dynamics predicted with Gaussian process regression. With Pln1-mNG a second, less 155 pronounced trough is visible late in the second half of the cell cycle (Figure 2A). In contrast to the area-156 normalised LD number, LD size is constant throughout the cell cycle (Figure 2C-D). Again, density plots 157 confirm the dynamics predicted with Gaussian process regression (Figure S3C-D) . The cell cycle 158 dynamics of the summed sizes of all LDs per cell, which notably is not normalised to the cell cross-area, 159 strongly resemble the dynamics of the area-normalised LD number (Figure S3E-F) . The similarity 160 between these dynamics indicates that the oscillation of the LD number is not an artefact resulting from 161 the normalisation to the cell cross-area. 162 We also noticed that the area-normalised LD number determined with mNG-Tgl3 is almost twofold 163 lower than th at measured with Pln1- mNG (Figure 2A- B). This difference could be explained by the 164 localisation of mNG-Tgl3 to a specific subset of LDs. First, when LD formation in a mutant strain is 165 induced by expression of the diacylglycerol acyltransferase Dga1 , Tgl3 is not detected on newly formed 166 LDs for the first hour after induction (Gao et al., 2017) . Second, since Tgl3 is an enzyme that uses TAG as 167 its substrate (Athenstaedt & Daum, 2003; Rajakumari & Daum, 2010) and subpopulations of TAG-168 specific LDs have been reported in mammalian cells (Hsieh et al., 2012) and S. pombe (Meyers et al., 169 2016), it is plausible that Tgl3 would localise only to TAG-enriched LDs, while Pln1 does not show this 170 specific localisation. 171 To investigate the similarities and differences of the oscillations observed with the two LD reporters, we 172 averaged the three replicates performed with each reporter , normalised the result to its mean and 173 plotted the normalised trajectories obtained with Pln1-mNG and mNG-Tgl3 against each other. Here, we 174 found a positive correlation from shortly before START until the final 15% of the cell cycle (Figure 2E), 175 indicating that the two markers report comparable dynamics of the area-normalised LD number during 176 this time interval. In contrast, the dynamics are different around mitotic exit. This difference could arise 177 from the different functions of the two LD reporter proteins. Specifically, since Pln1 is important for the 178 stabilisation of nascent LDs (Gao et al., 2017), the increase in area-normalised LD number observed with 179 Pln1-mNG at the end of the cell cycle could indicate the formation of new LDs . An increase in LD 180 numbers at the end of the cell cycle would be consistent with the above-average lipid biosynthetic 181 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 6 activity at the end of the cell cycle after its peak in the middle of S/G2/M (Takhaveev et al., 2023) and 182 the storage of triacylglycerol during septation (P. L. Yang et al., 2016). If neutral lipids are mobilised from 183 these same LDs after mitotic exit, leading to their disappearance, this would explain the more 184 pronounced drop in area-normalised LD number observed with Pln1-mNG compared to mNG-Tgl3. 185 Overall, employing two distinct LD reporter proteins, we have elucidated the general cell cycle dynamics 186 of LDs, which suggest that neutral lipids are mobilised from LDs between mitotic exit and START, as 187 evidenced by a trough in the area-normalised LD number, while the neutral lipid stores are replenished 188 during S/G2/M. These findings demonstrate that LDs are not stagnant organelles containing energy 189 reserves to be used in case of nutrient shortages. Instead, their content is mobilised and re-synthesised 190 as cells go through the cell cycle. 191 192 193 Figure 2. Number of lipid droplets oscillates over the cell cycle while LD size is constant. LDs were identified in time-lapse 194 microscopy images of cells expressing either Pln1- mNG (A, C) or mNG-Tgl3 (B , D) as an LD marker. For each LD reporter strain, 195 three biological replicates, whose results are represented by different line styles, were performed. Cell cycle trajectories were 196 aligned from one mitotic exit to the next (red vertical lines at cell cycle progression values 0 and 1) and for occurrence of START 197 (bright green vertical line) and budding (orange vertical line) between the se mitotic exit events. Gaussian process regression 198 was used to predict population averages of (A-B) area-normalised number of detected LDs and (C-D) size of the detected LDs; 199 (E) To compare the dynamics of area-normalised number of LDs as observed with the reporter proteins Pln1-mNG and mNG-200 Tgl3, we normalised every Gaussian process regression result to its own average value and plotted results obtained with mNG-201 Tgl3 against those obtained with Pln1-mNG. The grey line indicates y = x. Thin coloured lines denote the combination of 202 individual replicates using the two LD reporters (3x3 combinations) . The thick black line indicates combined results of the three 203 replicate experiments performed with each LD reporter and was obtained by averaging the three Gaussian process regression 204 outputs for every timepoint. The circular markers on the black curve denote the occurrence of mitotic exit, START and budding. 205 206 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 7 Dynamic LD fluorescence intensity likely results from LD number dynamics 207 After we discovered the cell cycle dynamics of the LD number, we asked whether the fluorescence 208 intensity measured on those LDs was also dynamic. Studying the dynamics of the fluorescence intensity 209 of the LD reporters Pln1-mNG and mNG-Tgl3 on LDs and in the cytoplasm could help us distinguish 210 between observations that reflect the general dynamics of LDs along the cell cycle and observations that 211 relate to the distinct biological functions of the LD marker proteins. We assessed the cell cycle dynamics 212 of the fluorescence intensity measured inside the whole cell mask, on LDs and in the cytoplasm. The 213 latter was determined by excluding all pixels belonging to an LD and calculating the average intensity of 214 all other pixels within the cell mask. While the fluorescence intensity inside the whole cell and in the 215 cytoplasm was found to be almost constant, the fluorescence intensity on LDs oscillates along the cell 216 cycle (Figure 3A-B) . With both Pln1-mNG and mNG-Tgl3, the fluorescence intensity on LDs peak ed 217 around START and was minimal during S/G2/M. 218 To compare the dynamics of LD fluorescence measured with the two reporters and to distinguish 219 between general and reporter protein-specific behaviour, we normalised the three replicates performed 220 with each marker protein to their respective means, took the average of the three replicates and plotted 221 the resulting trajectory for mNG-Tgl3 against that for Pln1-mNG . We found that the trajectories 222 obtained with the two LD marker proteins correlate positively (Figure 3C) , demonstrating that, along 223 most of the cell cycle, the oscillations of the fluorescence intensities measured with the two reporters 224 are comparable. However, this correlation is absent during the last stretch of the cell cycle, when the 225 fluorescence of mNG-Tgl3 on LDs is almost constant, while the fluorescence of Pln1-mNG on LDs is still 226 decreasing. The sharp peak of mNG-Tgl3 intensity on LDs around START could be related to its function 227 as a triacylglycerol lipase, as the decrease in the area-normalised LD number between mitotic exit and 228 START (Figure 2A-B) indeed indicates that neutral lipids are mobilised from LDs before START. 229 Comparing the dynamics of the area-normalised LD number (Figure 2A-B) and the LD fluorescence 230 intensity (Figure 3A-B), we noticed that the minimum of the area-normalised LD number around START 231 coincides with the maximum of the LD fluorescence intensity while, vice versa, the maximum of the 232 area-normalised LD number and the minimum in the fluorescence intensity on LDs both occur during 233 S/G2/M. Indeed, when we plotted the area-normalised LD number against the LD fluorescence, each cell 234 cycle trajectory normalised to its own mean, we saw an anticorrelation between these two 235 characteristics. This anticorrelation was observed both with Pln1-mNG and with mNG-Tgl3 as an LD 236 reporter protein (Figure 3D-E). 237 To comprehend why the fluorescence intensity of LDs would anticorrelate with the LD number, we 238 aimed to unite all observed LD characteristics, i.e. number, size and fluorescence intensity as well as the 239 concentration of the fluorescent reporter protein, in a unified explanation. First, we excluded that a 240 dynamic concentration of LD marker protein causes the cell cycle dynamics of the LD fluorescence 241 intensity. The concentration, proxied by the average fluorescence intensity within the cell mask, is 242 almost constant for both Pln1-mNG and mNG-Tgl3 (Figure 3A-B) and thus does not elicit the dynamic LD 243 fluorescence intensity. Second, changes in partitioning of the LD reporter proteins between LDs and the 244 cytoplasm cannot drive the changes in LD fluorescence either. If this partitioning changed, fluorescence 245 on LDs and in the cytoplasmic would show opposite trends, since marker protein moving from 246 cytoplasm to LDs would cause the cytoplasmic fluorescence intensity to decrease and the fluorescence 247 intensity of LDs to increase, and vice versa. However, the fluorescence intensity on LDs is dynamic, while 248 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 8 the fluorescence in the cytoplasm is constant (Figure 3A-B) . Third, we considered that LD size could 249 account for changes in the fluorescence intensity of LDs. If the amount of marker protein localised to an 250 LD stay ed constant while that LD increased in size, the marker protein would be dispersed and the 251 fluorescence intensity would decrease. Conversely, if the LD shrunk, the marker protein would become 252 more concentrated, increasing the fluorescence intensity. However, LD size is constant (Figure 2C-D), so 253 changes in LD size do not cause the dynamic fluorescence intensity of LDs. Lastly, the dynamic area-254 normalised LD number (Figure 2A-B) and its anticorrelation with the fluorescence intensity of LDs 255 (Figure 3D-E) can drive the changes in the LD fluorescence intensity. When the number of LDs is low, the 256 total pool of reporter protein is spread across few LDs, resulting in high fluorescence intensities on those 257 LDs. When the number of LDs increases, the same reporter protein molecules are distributed over more 258 LDs, and consequently, the fluorescence intensity on the LDs will be lower. Thus, the observed changes 259 in LD number could be responsible for the dynamics in LD fluorescence intensity , whereby the 260 anticorrelation between fluorescence intensity of LDs and the area-normalised LD number would be 261 explained. 262 Overall, our results show that the fluorescence intensity of LDs oscillates along the cell cycle and 263 anticorrelates with the area-normalised number of LDs. In contrast, reporter protein concentrations, the 264 partitioning of reporter proteins between LDs and the cytoplasm, and LD size are constant. Together, 265 these findings indicate that the cell cycle dynamics of the LD fluorescence intensity are likely driven by 266 the oscillating LD number. 267 268 269 Figure 3. Dynamic LD fluorescence intensity anticorrelates with area-normalised LD number. LDs were identified in time-lapse 270 microscopy images of cells expressing either Pln1-mNG (A, D) or mNG-Tgl3 (B, E) as an LD reporter protein. For each reporter 271 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 9 strain, multiple biological replicates, whose results are represented by different line styles, were performed. Cell cycle 272 trajectories were aligned from one mitotic exit to the next (red vertical lines at cell cycle progression values 0 and 1) and for the 273 occurrence of START (bright green vertical line) and budding (orange vertical line) between the two subsequent instances of 274 mitotic exit. Gaussian process regression was used to predict population averages of (A-B) average fluorescence intensity of 275 detected LDs, the cytoplasmic region and the whole cell mask; (C) To allow direct comparison between the cell cycle dynamics 276 of the fluorescence intensity of LDs detected with Pln1-mNG or mNG-Tgl3, we normalised every Gaussian process regression 277 output to its own mean value and plotted the trajectories obtained with mNG-Tgl3 against those obtained with Pln1-mNG. The 278 grey line indicates y = x, on which all data points would lie in case of perfect correlation. Thin coloured lines denote 279 combinations of individual replicates (3x3 combinations). The thick black line represents the averaged results of the three 280 replicates performed with each LD reporter protein. The circular markers on this curve represent the occurrence of the cell 281 cycle events mitotic exit, START and budding; (D-E) To show anticorrelation between the area-normalised LD number and the 282 fluorescence intensity measured on LDs, we normalised every Gaussian process output to its own mean and plotted the results 283 against each other. Thin coloured lines denote the results for the individual replicates and the thick black line represent 284 averaged results of the three replicates; circular markers on this black trajectory denote mitotic exit, START and budding. The 285 grey line indicates where data points would lie in case of perfect anticorrelation. 286 287 TAG storage and mobilisation give rise to LD dynamics 288 After we discovered the cell cycle dynamics of LD number and fluorescence intensity, we wondered 289 whether we could identify which biological process is responsible for the oscillations. Changes in the LD 290 number and fluorescence could be due to changing synthesis and mobilisation of TAG and steryl esters , 291 as well as fission and fusion of LDs. To test whether TAG metabolism was responsible for the LD cell 292 cycle dynamics, we perturb ed TAG metabolism and subsequently observed the area-normalised LD 293 number along the cell cycle. To perturb TAG synthesis, we delet ed the genes encoding the two major 294 TAG synthases in S. cerevisiae Lro1 (Oelkers et al., 2000) and Dga1 (Oelkers et al., 2002). In a separate 295 strain, we deleted the genes encoding the TAG lipases Tgl3 (Athenstaedt & Daum, 2003) and Tgl4 296 (Athenstaedt & Daum, 2005), thereby blocking the mobilisation of TAG from LDs. Notably, in the cells 297 with perturbed TAG metabolism, we only used Pln1-mNG as an LD reporter, since our other reporter 298 protein, Tgl3, was deleted in one of the TAG mutant strains. 299 Before investigating the LD cell cycle dynamics in the two double deletion strains, we first assessed how 300 the perturbation of TAG metabolism affect ed the LD phenotype independent of the cell cycle stage. To 301 this end, we analysed fluorescence microscopy snapshots of cells from exponential cultures. W e saw 302 that fluorescence intensity of Pln1-mNG both within the entire cell mask and on LDs was higher in 303 ΔTGL4ΔTGL3 compared to the wild type, and lower in ΔDGA1ΔLRO1 (Figure S4). These changes in the 304 expression of Pln1, which is important for the formation and stabilisation of LDs, imply that the number 305 and size of LDs could also be different in the two mutation strains . Indeed, we found that the area-306 normalised LD number was significantly lower in both deletion strains relative to the wild type (Figure 307 4A), with no puncta detected in 44% of ΔDGA1ΔLRO1 cells and 11% of ΔTGL3ΔTGL4 cells. Moreover, the 308 LDs in both deletion strains were bigger than those in the wild type (Figure 4B). Therefore, perturbation 309 of TAG metabolism, preventing its synthesis or mobilisation, affects both the number of LDs and their 310 size. 311 Some of the changes in the LD phenotype may at first glance seem counterintuitive, but are in line with 312 previous research. In t he ΔDGA1ΔLRO1 mutant, which is unable to synthesise TAG, we found larger LDs 313 than in the wild type (Figure 4B) while one may expect a decrease in LD size due to the absence of TAG. 314 The stabilisation of specifically small LDs by diacylglycerol acyltransferases (Kovacs et al., 2021; Wilfling 315 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 10 et al., 2013) could explain the increased size of LDs in ΔDGA1ΔLRO1. The other deletion strain, 316 ΔTGL3ΔTGL4 can synthesise TAG, but cannot mobilise it from LDs. Therefore, o ne would not expect the 317 observed decrease in LD number compared to the wild type (Figure 4A) . As Tgl4 is involved in the 318 stabilisation of nascent LDs (Wang et al., 2024), its absence in ΔTGL3ΔTGL4 may hinder LD formation, 319 resulting in lower LD numbers compared to the wild type. Overall, these findings show that deleting the 320 genes encoding the enzymes that synthetise or mobilise TAG changes LD morphology, and can have 321 counterintuitive effects due to additional functions of these enzymes. 322 Next, we performed time-lapse microscopy experiments to investigate the cell cycle dynamics of LDs in 323 the deletion strains and thereby determine whether TAG metabolism contributes to LD dynamics. In 324 both deletion backgrounds, the cell cycle oscillations of the area-normalised LD number as observed in 325 the wild type were lost (Figure 4C). Similarly to the wild type, LD size was constant along the cell cycle in 326 both deletion backgrounds (Figure 4D). As LD fission would lead to the appearance of two smaller LDs 327 originating from one larger LD, while LD fusion would have the opposite effect, the constant LD size 328 along the cell cycle makes it improbable that LD fission and fusion contribute to the cell cycle dynamics 329 of the LD number. In contrast, the constant area-normalised LD number in mutants unable to synthesise 330 or mobilise TAG suggests that TAG metabolism must give rise to the LD dynamics as observed in wild-331 type cells. Hereby, we have established that TAG metabolism, instead of the storage and mobilisation of 332 steryl esters or the fission and fusion of LDs, underlies the oscillations of the LD number during the cell 333 cycle. 334 335 336 Figure 4. TAG storage and mobilisation give rise to LD dynamics . (A-B) Area-normalised number of LDs and LD size as 337 determined from snapshots of cells from exponential cultures that express Pln1-mNG as an LD reporter in wild type, 338 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 11 ΔDGA1ΔLRO1 and ΔTGL3ΔTGL4 (298, 314 and 264 cells, respectively). The median is indicated with a white diamond and an 339 open circle indicates the mean. In both deletion backgrounds, the number of detected LDs per cell cross-area is significantly 340 lower and LDs are significantly larger than in the wild type (two-sided Mann-Whitney U-test; p < 0.05). The percentage of cells 341 without any detected LDs is indicated next to each violin in A; (C-D) Cell cycle dynamics of the area-normalised L D number and 342 size of detected LDs predicted with Gaussian process regression applied to cell cycle-aligned single-cell trajectories from three 343 biological backgrounds, indicated in different colours. Biological replicates are indicated by different line styles. Cell cycles were 344 aligned from one occurrence of mitotic exit to the next (red vertical lines at cell cycle progression values 0 and 1) and for 345 occurrence of START (solid vertical lines) and budding (dashed vertical lines). 346 347 Perturbing LDs through TAG metabolism delays START 348 Finally, we asked if the LD dynamics, as observed in the wild type but lost in the strains with perturbed 349 TAG metabolism, could affect cell cycle progression. Since START is delayed when cells that cannot 350 mobilise TAG from LDs resume growth after starvation (Kurat et al., 2009), we wondered if the same 351 could be true in exponentially growing cells. To investigate this, we assessed the interrelation between 352 the duration of the whole cell cycle and the time between mitotic exit and START in individual cell 353 cycles. We plotted the duration of the mitotic exit to START phase against the whole cell cycle length in 354 ΔDGA1ΔLRO1, ΔTGL3ΔTGL4 and the wild type and performed linear regression to obtain equations that 355 describe the ir interrelation (Figure 5A) . The regression lines describe the duration of mitotic exit to 356 START as a function of cell cycle length and therefore, their slopes indicate the fraction of the cell cycle 357 taken up by mitotic exit to START. We found that the slopes obtained from both deletion strains were 358 steeper than those from the wild type (Figure 5B), which means that mitotic exit to START takes up a 359 larger fraction of the cell cycle in TAG mutants compared to the wild type. The change in the fraction of 360 the cell cycle taken up by mitotic exit to START detected on the single-cell cycle level was not 361 accompanied by extensive population-level changes in absolute duration of the cell cycle or its phases 362 mitotic exit to START, START to budding and budding to mitotic exit (Figure S5). The absolute changes in 363 duration between the wild type and the two TAG mutants were comparable for the cell cycle phases and 364 the whole cell cycle and on average were equal to 5 minutes, which corresponds to one imaging 365 interval. Together , these results show that in cell cycles of identical length, START occurs later in 366 ΔDGA1ΔLRO1 and ΔTGL3ΔTGL4 than in the wild type. This suggests that START is delayed in cells that 367 cannot synthesise or mobilise TAG, which signifies that intact TAG metabolism is important for the 368 timely occurrence of START. 369 370 371 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 12 Figure 5. Perturbing LDs through TAG metabolism delays START. (A) Interrelation between the duration of mitotic exit to START 372 and duration of the whole cell cycle in the wild type, ΔDGA1ΔLRO1 and ΔTGL3ΔTGL4. Different colours represent replicate 373 experiments and marker size scales with the number of times a combination of cell cycle length and mitotic exit to START 374 duration was observed. Trend lines obtained with linear regression describe the duration of mitotic exit to START duration as a 375 function of cell cycle length ; (B) Variation in the slope values of the regression lines from A, estimated with bootstrapping. A 376 total of 100 bootstrapping iterations were performed for every experiment. In each iteration, 50% of data points were 377 randomly sampled with replacement and subsequently, linear regression was performed to obtain a slope value. Mean and 378 standard deviation of the slope values obtained with bootstrapping are shown in black. Grey horizontal lines indicate the slo pe 379 values of the regression lines in A, which were obtained using all data points. 380 381

382 In this work, we established Pln1, Tgl3 and Rrt8 as marker proteins for LDs by showing their 383 colocalisation with LDs stained with BODIPY-TR. Using Pln1-mNeonGreen and mNeonGreen-Tgl3 as LD 384 reporters in time-lapse microscopy experiments, we found that the ar ea-normalised LD number 385 oscillates during the cell cycle, with a minimum around START and a peak halfway through S/G2/M . We 386 perturbed TAG metabolism by deleting the genes that encode the acyl transferases Dga1 and Lro1 or the 387 lipases Tgl3 and Tgl4, respectively preventing the synthesis of TAG or its mobilisation from LDs. Both sets 388 of gene deletions abolished the cell cycle dynamics in the area-normalised LD number. Furthermore, in 389 cell cycles of identical duration, START on average was delayed in the two double deletion backgrounds 390 compared to the wild type , suggesting that the mobilisation of TAGs from LDs is important for START to 391 take place. Thus, our results demonstrate the importance of storage lipid metabolism for cell cycle 392 progression. 393 The LD cell cycle dynamics that we observed are supported by a number of previous publications, 394 indicating biological processes that m ay contribute to the dynamic behaviour of LDs . First, cells that re-395 enter the cell cycle after starvation mobilise neutral lipids from LDs (Kurat et al., 2006; Rajakumari & 396 Daum, 2010). As we have shown here, exponentially growing cells do the same when passing START. 397 Furthermore, both the protein levels of the fatty acid synthesis enzymes Acc1, Fas1 and Fas2 (Blank et 398 al., 2017) and lipid biosynthetic activity (Takhaveev et al., 2023) peak in S/G2/M, which could explain the 399 increase in area-normalised LD number we observed during this part of the cell cycle. Moreover, the 400 lipids that are mobilised from LDs during the second half of S/G2/M could partake in triglyceride cycling, 401 a process in which TAG is partially degraded and then re-synthesised with different fatty acid chains, to 402 metabolically alter stored neutral lipids and change their exact molecular identity (Wunderling et al., 403 2023). Finally, neutral lipid storage is important right before mitotic exit, as cells that lack all LDs exhibit 404 cytokinetic defects, which are rescued by chemical inhibition of lipid synthesis (P. L. Yang et al., 2016). 405 However, a recent study that quantified TAG in cells going through the yeast metabolic cycle (YMC) (S. 406 Yang et al., 2025) contradicts the neutral lipid storage dynamics inferred from the LD number in the 407 current work. Yang et al. show an increase in TAG levels early in the YMC, followed by a gradual 408 decrease that lasts until the cycle is completed. We believe that the discrepancy between the LD 409 dynamics we report and these TAG dynamics along the YMC could be due to only 40% of these 410 metabolically synchronised cells dividing. Since LDs are dynamic during stationary phase (Hariri et al., 411 2018; Qiu et al., 2023; Wang et al., 2014) when cells no longer divide, the dividing and non-dividing cells 412 may well have distinct TAG dynamics . In this case, the reported TAG dynamics along the YMC would not 413 match the cell cycle dynamics of TAG, which in turn could explain the differences between the dynamics 414 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 13 of TAG (S. Yang et al., 2025) and the cell cycle dynamics of the area-normalised LD number reported in 415 the current work. 416 Multiple cell cycle regulators could be involved in the orchestration of LD dynamics. The cyclin-417 dependent kinase Cdc28 activates the lipase Tgl4 (Kurat et al., 2009) and inhibits Pah1 (Choi et al., 2011; 418 Santos-Rosa et al., 2005), which synthesises th e TAG precursor diacylglycerol. Pah1 is also regulated by 419 PKA (Su et al., 2018) and PKC (Su et al., 2014) . Activity of TORC1 along the cell cycle (Guerra, 420 Vuillemenot, Van Oppen, et al., 2022) coincid es with periods of mobilisation of neutral lipids from LDs . 421 Indeed, TORC1 activity has been shown to stimulate the mobilisation of neutral lipids from LDs (Madeira 422 et al., 2014). Thus, the cell cycle regulators Cdc28, PKA, PKC and TORC1 conceivably could participate in 423 the coordination of the cell cycle dynamics of LDs. 424 We have shown that mobilisation of TAG from LDs contributes to the timely occurrence of START. 425 Previous findings indicate that the mobilised lipids could be used as precursors for sphingolipid 426 synthesis. Supplementation with a sphingolipid precursor rescues the delay in START in ΔTGL3ΔTGL4 427 (Chauhan et al., 2015) while inhibition of sphingolipid synthesis inhibits the G1/S transition (Cerbón et 428 al., 2005). Sphingolipids can activate the phosphatase PP2A Cdc55 (Nickels & Broach, 1996), which in turn 429 promotes START (McCourt et al., 2013; Moreno-Torres et al., 2015). Thus, sphingolipid synthesis can 430 drive cell cycle progression and lack thereof could explain the delay in START observed in TAG mutant 431 strains. 432 Overall, we have discovered that the number of LDs oscillates along the cell cycle , that this oscillation 433 depends on TAG metabolism and that the mobilisation of TAG from LDs is important for the timely 434 occurrence of START. Our findings highlight the importance of storage lipid metabolism in cell cycle 435 progression and emphasise that LD storage should be considered in research centred on commitment to 436 the cell cycle. Further research is still needed to elucidate the regulatory processes underlying the 437 observed dynamics of LDs. 438 439

440 Strains 441 The yeast strains used in this study (Table S1) were constructed from the wild-type strain YSBN6 (S288C 442 background) (Canelas et al., 2010) or from YSBN6 WHI5::mCherry-BLE (Litsios et al., 2019), using a 443 CRISPR-Cas9 based approach (Novarina et al., 2022) with primers listed in Table S2. To tag proteins with 444 the fluorescent protein mNeonGreen, we used a codon-optimised sequence for expression in S. 445 cerevisiae (Guerra, Vuillemenot, Rae, et al., 2 022). Gene deletions were verified by PCR. Introduction of 446 a fluorescent protein on target proteins was verified by PCR and sequencing. 447 Culturing 448 Cells were grown in 10 mL of Verduyn minimal medium (Verduyn et al., 1992) buffered at pH 5.0 with 10 449 mM potassium phthalate and with 2% glucose as a carbon source in 100 mL flasks at 30°C, under 450 constant rotation at 300 rpm. To obtain snapshots of live cells, 2-4 µL of cells were taken from an 451 exponentially growing culture and placed on a microscopy slide under a 1% agarose pad soaked with the 452 culture medium. To prepare cells for microfluidic experiments, cultures were maintained in the 453 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 14 exponential phase for at least 12 h prior to the experiment through repeated dilution. On the day of the 454 experiment, cultures were diluted to an OD600 of 0.05 and after one additional doubling, cells were 455 loaded into the microfluidic chip. Microfluidics experiments were performed as described (Huberts et 456 al., 2013). The flow rate during microfluidics experiments was 4 µL/min. 457 Formaldehyde fixation and BODIPY-TR staining 458 Cells were grown for 24 h to reach stationary phase. The equivalent of 1 mL of a culture with an OD 600 of 459 5 was harvested by centrifugation (3 min, 10 000 g). Next, cells were fixed with formaldehyde as 460 described (Madeira et al., 2015). Briefly, cells were resuspended in 1 mL of 3.7% (w/v) 461 paraformaldehyde with 0.1 M sorbitol as an osmolyte and incubated at room temperature for 15 min. 462 Afterwards, cells were washed once with 1 mL of phosphate buffered saline (PBS) and finally 463 resuspended in 1 mL of PBS. 464 To stain formaldehyde-fixed cells with BODIPY-TR methyl ester (Lumiprobe GmbH) 1 µL of 5 mM 465 BODIPY-TR methyl ester dissolved in DMSO was added to 100 µL of fixed cell suspension and incubated 466 at room temperature for 5 min. Cells were then pelleted (1.5 min, 12 500 g), washed once with 100 µL of 467 PBS and finally resuspended in 100 µL of PBS. Cells were imaged immediately after staining. 468 Microscopy 469 All images were acquired on a Nikon Eclipse Ti-E inverted wide-field fluorescence microscope equipped 470 with the Nikon Perfect Focus System (PFS) and an Andor-DU -897 EX camera. For imaging, a 100x S Fluor 471 Oil objective (NA 1. 4, Nikon) was used and the readout mode was set to 1 MHz without gain 472 amplification. For bright-field images, excitation was done with a halogen lamp fitted with a 420 nm 473 beam-splitter to filter out short wavelengths. A Lumencor AURA excitation system was used for 474 fluorescence excitation. Green fluorescent proteins were excited at 485 nm with an imaging set-up 475 consisting of a 470/40 nm band-pass filter, a 495 nm beam splitter and a 525/50 nm emission filter. Red 476 fluorophores were imaged using excitation at 565 nm, a 560/40 nm band-pass filter, a 585 nm beam 477 splitter and a 630 nm/75 nm emission filter. Intensity of the light source and excitation time for each 478 fluorophore and each experiment are detailed in Table S3. For live-cell imaging, the microscope setup 479 was kept at 30 °C using an incubator box (Life Imaging Services). 480 Optimising image alignment 481 To correct the small-scale but systematic misalignment between images recorded in the GFP channel 482 (mNeonGreen-tagged proteins) and the RFP channel (BODIPY-TR), we used the bilinear transformation 483 approach from the Python module pyStackReg (Thévenaz et al., 1998) . To obtain the transformation 484 matrix, we created composite images of binarised snapshot images of cells that express Pln1-485 mNeonGreen and that were also stained with the red fluorophore BODIPY-TR. Binarisation of the 486 snapshot images ensured that only outlines of the cells, but no cytoplasmic structures, were visible. 487 Combination of multiple snapshot images yielded a composite image showing the outline of cells 488 dispersed over the whole field of view. We aligned the composite of RFP snapshots to the composite of 489 the corresponding GFP snapshots, thereby obtaining a transformation matrix which we used to 490 optimally align images recorded in the RFP channel to images recorded in the GFP channel. 491 Cell segmentation and cell cycle alignment 492 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 15 Fluorescence images were background corrected using the rolling ball algorithm implemented in ImageJ 493 with a radius of 50 pixels. Cell masks were obtained from bright-field images with the ImageJ plugin BudJ 494 (Ferrezuelo et al., 2012). Segmentation was inspected visually to verify that the cell masks match ed the 495 cells in the bright-field images. Finally, cell cycles were excluded from further analysis if their total 496 duration, from one mitotic exit to the next, exceeded 180 min. For all cells, we manually tracked mitotic 497 exit, budding and START. Budding was detected in the bright-field images, START was detected from 498 Whi5-mCherry leaving the nucleus and mitotic exit from Whi5-mCherry entering the nucleus. 499 For the cell cycles that passed the selection criteria ( i.e. proper segmentation and cell cycle duration), 500 we transformed the data onto a common cell cycle progression coordinate ranging from 0 to 1 to allow 501 comparison between cycles of different durations. Consecutive mitotic exits were defined as 0 and 1. 502 We also determined the average position of START and budding on this common time coordinate. To do 503 so, we calculated the fraction of the cell cycle that had passed at the moment that each event occurred. 504 Specifically, to determine the timing of START, we divided the time between the first mitotic exit and 505 START by the duration of the whole cell cycle. We repeated this procedure for budding. Ultimately, 506 combining data from all cell cycles, we determined the average timing of START and budding on the 507 normalised cell cycle progression coordinate. 508 Subsequently, we determined the cell cycle progression coordinate value for every data point recorded 509 in the time-lapse microscopy experiments. To do this, we divided each cycle into three phases: mitotic 510 exit to START, START to budding and budding to mitotic exit. For each of these phases, we placed its first 511 and last data point, which coincide with a cell cycle event, at the normalised time values for these 512 events. Then, we evenly dispersed all interstitial data points over the time interval bounded by the two 513 cell cycle events. Performing this procedure for every cell cycle, we determined a time value on the 514 common cell cycle progression coordinate for every recorded data point. 515 To infer the population-average behaviour of the various measures over the cell cycle, we performed 516 Gaussian process regression of the cell cycle aligned single-cell data. For this, we used Python’s 517 sklearn.gaussian_process (Pedregosa et al., 2011) using the radial basis function (RBF) as a 518 prior, with the length scale range [0.01, 0.5] and the white kernel with free noise level. An optimised fit 519 was obtained through maximisation of the log-marginal likelihood in the regression. 520 Automated detection of LDs in fluorescence microscopy images 521 We used the PunctaFinder algorithm (Terpstra et al., 2024) to automatically detect LD puncta and 522 estimate their size, both for LDs stained with BODIPY-TR in the RFP channel and for LDs visualised with 523 fluorescently tagged marker proteins in the GFP channel. We used an overlap parameter value of zero 524 and a punctum diameter of three pixels, which reflects a distance of 480 nm on our microscopy setup 525 and corresponds to the diameter of an average lipid droplet ( i.e. 400 nm) (Czabany et al., 2008). To 526 obtain suitable threshold values for punctum detection, we created a manually curated data set of 40 527 cells for each genetic background and performed threshold value optimisation with five bootstrap 528 iterations, sampling 75% of cells with replacement. Each final threshold value was the average of the 529 values obtained in the five bootstrap iterations. The threshold values are provided in Table S4. 530 Colocalisation analysis 531 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 16 Colocalisation between puncta of LDs stained with BODIPY-TR and puncta of LD marker proteins tagged 532 with mNeonGreen was assessed based on the distance between the midpoints of puncta in the two 533 imaging channels. Puncta are qualified as colocalising if their midpoints are maximally one pixel apart in 534 both the x- and y- direction. Colocalisation is qualified as ambiguous for puncta whose midpoints are 535 two pixels apart in one direction, but maximally one pixel apart in the other. In all other cases, 536 colocalisation is qualified as non-existent. 537

538 Athenstaedt, K., & Daum, G. (2003). YMR313c/TGL3 encodes a novel triacylglycerol lipase located in lipid 539 particles of Saccharomyces cerevisiae. Journal of Biological Chemistry, 278(26), 23317–23323. 540 https://doi.org/10.1074/jbc.M302577200 541 Athenstaedt, K., & Daum, G. (2005). Tgl4p and Tgl5p, Two Triacylglycerol Lipases of the Yeast 542 Saccharomyces cerevisiae Are Localized to Lipid Particles *. Journal of Biological Chemistry, 543 280(45), 37301–37309. https://doi.org/10.1074/JBC.M507261200 544 Blank, H. M., Papoulas, O., Maitra, N., Garge, R., Kennedy, B. K., Schilling, B., Marcotte, E. M., & 545 Polymenis, M. (2020). Abundances of transcripts, proteins, and metabolites in the cell cycle of 546 budding yeast reveal coordinate control of lipid metabolism. Molecular Biology of the Cell, 31(10), 547 1069–1084. https://doi.org/10.1091/MBC.E19-12-0708/ASSET/IMAGES/LARGE/MBC-31-1069-548 G006.JPEG 549 Blank, H. M., Perez, R., He, C., Maitra, N., Metz, R., Hill, J., Lin, Y., Johnson, C. D., Bankaitis, V. A., 550 Kennedy, B. K., Aramayo, R., & Polymenis, M. (2017). Translational control of lipogenic enzymes in 551 the cell cycle of synchronous, growing yeast cells. The EMBO Journal, 36(4), 487–502. 552 https://doi.org/10.15252/EMBJ.201695050 553 Breker, M., Gymrek, M., & Schuldiner, M. (2013). A novel single-cell screening platform reveals 554 proteome plasticity during yeast stress responses. The Journal of Cell Biology, 200(6), 839–850. 555 https://doi.org/10.1083/JCB.201301120 556 Canelas, A. B., Harrison, N., Fazio, A., Zhang, J., Pitkänen, J. P., van den Brink, J., Bakker, B. M., Bogner, L., 557 Bouwman, J., Castrillo, J. I., Cankorur, A., Chumnanpuen, P., Daran-Lapujade, P., Dikicioglu, D., van 558 Eunen, K., Ewald, J. C., Heijnen, J. J., Kirdar, B., Mattila, I., … Nielsen, J. (2010). Integrated 559 multilaboratory systems biology reveals differences in protein metabolism between two reference 560 yeast strains. Nature Communications, 1(9). https://doi.org/10.1038/NCOMMS1150 561 Cerbón, J., Falcon, A., Hernández-Luna, C., & Segura-Cobos, D. (2005). Inositol phosphoceramide 562 synthase is a regulator of intracellular levels of diacylglycerol and ceramide during the G1 to S 563 transition in Saccharomyces cerevisiae. Biochemical Journal, 388(1), 169–176. 564 https://doi.org/10.1042/BJ20040475 565 Chauhan, N., Visram, M., Cristobal-Sarramian, A., Sarkleti, F., & Kohlwein, S. D. (2015). Morphogenesis 566 checkpoint kinase Swe1 is the executor of lipolysis-dependent cell-cycle progression. Proceedings 567 of the National Academy of Sciences of the United States of America, 112(10), E1077–E1085. 568 https://doi.org/10.1073/PNAS.1423175112/SUPPL_FILE/PNAS.201423175SI.PDF 569 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 17 Choi, H. S., Su, W. M., Morgan, J. M., Han, G. S., Xu, Z., Karanasios, E., Siniossoglou, S., & Carman, G. M. 570 (2011). Phosphorylation of Phosphatidate Phosphatase Regulates Its Membrane Association and 571 Physiological Functions in Saccharomyces cerevisiae: IDENTIFICATION OF SER602, THR723, AND 572 SER744 AS THE SITES PHOSPHORYLATED BY CDC28 (CDK1)-ENCODED CYCLIN-DEPENDENT KINASE 573 *. Journal of Biological Chemistry, 286(2), 1486–1498. https://doi.org/10.1074/JBC.M110.155598 574 Chong, Y. T., Koh, J. L. Y., Friesen, H., Duffy, K., Cox, M. J., Moses, A., Moffat, J., Boone, C., & Andrews, B. 575 J. (2015). Yeast Proteome Dynamics from Single Cell Imaging and Automated Analysis. Cell, 161(6), 576 1413–1424. https://doi.org/10.1016/J.CELL.2015.04.051 577 Czabany, T., Wagner, A., Zweytick, D., Lohner, K., Leitner, E., Ingolic, E., & Daum, G. (2008). Structural 578 and Biochemical Properties of Lipid Particles from the Yeast Saccharomyces cerevisiae *. Journal of 579 Biological Chemistry, 283, 17065–17074. https://doi.org/10.1074/jbc.M800401200 580 Eisenberg-Bord, M., Mari, M., Weill, U., Rosenfeld-Gur, E., Moldavski, O., Castro, I. G., Soni, K. G., 581 Harpaz, N., Levine, T. P., Futerman, A. H., Reggiori, F., Bankaitis, V. A., Schuldiner, M., & Bohnert, 582 M. (2018). Identification of seipin-linked factors that act as determinants of a lipid droplet 583 subpopulation. The Journal of Cell Biology, 217(1), 269. https://doi.org/10.1083/JCB.201704122 584 Ferrezuelo, F., Colomina, N., Palmisano, A., Garí, E., Gallego, C., Csikász-Nagy, A., & Aldea, M. (2012). The 585 critical size is set at a single-cell level by growth rate to attain homeostasis and adaptation. Nature 586 Communications, 3, 1012. https://doi.org/10.1038/ncomms2015 587 Ganesan, S., Tavassoli, M., Shabits, B. N., & Zaremberg, V. (2020). Tubular ER Associates With 588 Diacylglycerol-Rich Structures During Lipid Droplet Consumption. Frontiers in Cell and 589 Developmental Biology, 8. https://doi.org/10.3389/FCELL.2020.00700/FULL 590 Gao, Q., Binns, D. D., Kinch, L. N., Grishin, N. V., Ortiz, N., Chen, X., & Goodman, J. M. (2017). Pet10p is a 591 yeast perilipin that stabilizes lipid droplets and promotes their assembly. The Journal of Cell 592 Biology, 216(10), 3199. https://doi.org/10.1083/JCB.201610013 593 Grillitsch, K., Connerth, M., Köfeler, H., Arrey, T. N., Rietschel, B., Wagner, B., Karas, M., & Daum, G. 594 (2011). Lipid particles/droplets of the yeast Saccharomyces cerevisiae revisited: Lipidome meets 595 Proteome. Biochimica et Biophysica Acta - Molecular and Cell Biology of Lipids, 1811(12), 1165–596 1176. https://doi.org/10.1016/j.bbalip.2011.07.015 597 Guerra, P., Vuillemenot, L. A., Rae, B., Ladyhina, V., & Milias-Argeitis, A. (2022). Systematic in Vivo 598 Characterization of Fluorescent Protein Maturation in Budding Yeast. ACS Synthetic Biology, 11(3), 599 1129–1141. https://doi.org/10.1021/ACSSYNBIO.1C00387/SUPPL_FILE/SB1C00387_SI_002.ZIP 600 Guerra, P., Vuillemenot, L.-A. P. E., Van Oppen, Y. B., Been, M., & Milias-Argeitis, A. (2022). TORC1 and 601 PKA activity towards ribosome biogenesis oscillates in synchrony with the budding yeast cell cycle. 602 Journal of Cell Sciences, 135(18). https://doi.org/10.1242/jcs.260378 603 Hariri, H., Rogers, S., Ugrankar, R., Liu, Y. L., Feathers, J. R., & Henne, W. M. (2018). Lipid droplet 604 biogenesis is spatially coordinated at ER –vacuole contacts under nutritional stress . EMBO Reports, 605 19(1), 57–72. https://doi.org/10.15252/EMBR.201744815/SUPPL_FILE/EMBR201744815-SUP-606 0004-MOVIEEV2.ZIP 607 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 18 Hsieh, K., Lee, Y. K., Londos, C., Raaka, B. M., Dalen, K. T., & Kimmel, A. R. (2012). Perilipin family 608 members preferentially sequester to either triacylglycerol-specific or cholesteryl-esterspecific 609 intracellular lipid storage droplets. Journal of Cell Science, 125(17), 4067–4076. 610 https://doi.org/10.1242/JCS.104943/263163/AM/PERILIPIN-FAMILY-MEMBERS-PREFERENTIALLY-611 SEQUESTER 612 Huberts, D. H. E. W., Sik Lee, S., Gonzáles, J., Janssens, G. E., Vizcarra, I. A., & Heinemann, M. (2013). 613 Construction and use of a microfluidic dissection platform for long-term imaging of cellular 614 processes in budding yeast. Nature Protocols, 8(6), 1019–1027. 615 https://doi.org/10.1038/nprot.2013.060 616 Huh, W. K., Falvo, J. V., Gerke, L. C., Carroll, A. S., Howson, R. W., Weissman, J. S., & O’Shea, E. K. (2003). 617 Global analysis of protein localization in budding yeast. Nature, 425(6959), 686–691. 618 https://doi.org/10.1038/nature02026 619 Kovacs, M., Geltinger, F., Verwanger, T., Weiss, R., Richter, K., & Rinnerthaler, M. (2021). Lipid Droplets 620 Protect Aging Mitochondria and Thus Promote Lifespan in Yeast Cells. Frontiers in Cell and 621 Developmental Biology, 9, 3227. https://doi.org/10.3389/FCELL.2021.774985/BIBTEX 622 Kurat, C. F., Natter, K., Petschnigg, J., Wolinski, H., Scheuringer, K., Scholz, H., Zimmermann, R., Leber, R., 623 Zechner, R., & Kohlwein, S. D. (2006). Obese Yeast: Triglyceride Lipolysis Is Functionally Conserved 624 from Mammals to Yeast. Journal of Biological Chemistry, 281(1), 491–500. 625 https://doi.org/10.1074/JBC.M508414200 626 Kurat, C. F., Wolinski, H., Petschnigg, J., Kaluarachchi, S., Andrews, B., Natter, K., & Kohlwein, S. D. 627 (2009). Cdk1/Cdc28-Dependent Activation of the Major Triacylglycerol Lipase Tgl4 in Yeast Links 628 Lipolysis to Cell-Cycle Progression. Molecular Cell, 33(1), 53–63. 629 https://doi.org/10.1016/J.MOLCEL.2008.12.019 630 Lee, S. S., Avalos Vizcarra, I., Huberts, D. H. E. W., Lee, L. P., & Heinemann, M. (2012). Whole lifespan 631 microscopic observation of budding yeast aging through a microfluidic dissection platform. 632 Proceedings of the National Academy of Sciences of the United States of America, 109(13), 4916–633 4920. https://doi.org/10.1073/pnas.1113505109 634 Lin, C. P.-C., Kim, C., Smith, S. O., & Neiman, A. M. (2013). A Highly Redundant Gene Network Controls 635 Assembly of the Outer Spore Wall in S. cerevisiae. PLoS Genetics, 9(8), e1003700. 636 https://doi.org/10.1371/journal.pgen.1003700 637 Litsios, A., Huberts, D. H. E. W., Terpstra, H. M., Guerra, P., Schmidt, A., Buczak, K., Papagiannakis, A., 638 Rovetta, M., Hekelaar, J., Hubmann, G., Exterkate, M., Milias-Argeitis, A., & Heinemann, M. (2019). 639 Differential scaling between G1 protein production and cell size dynamics promotes commitment 640 to the cell division cycle in budding yeast. Nature Cell Biology, 21(11), 1382–1392. 641 https://doi.org/10.1038/S41556-019-0413-3 642 Madeira, J. B., Masuda, C. A., Maya-Monteiro, C. M., Matos, G. S., Montero-Lomelí, M., & Bozaquel-643 Morais, B. L. (2014). TORC1 inhibition induces lipid droplet replenishment in yeast. Molecular and 644 Cellular Biology, 35(4), 737–746. https://doi.org/10.1128/MCB.01314-14 645 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 19 Madeira, J. B., Masuda, C. A., Maya-Monteiro, C. M., Soares Matos, G., Montero-Lomelí, M., & Bozaquel-646 Morais, B. L. (2015). TORC1 Inhibition Induces Lipid Droplet Replenishment in Yeast. 647 https://doi.org/10.1128/MCB.01314-14 648 Markgraf, D. F., Klemm, R. W., Junker, M., Hannibal-Bach, H. K., Ejsing, C. S., & Rapoport, T. A. (2014). An 649 ER protein functionally couples neutral lipid metabolism on lipid droplets to membrane lipid 650 synthesis in the ER. Cell Reports, 6(1), 44–55. https://doi.org/10.1016/J.CELREP.2013.11.046 651 McCourt, P., Gallo-Ebert, C., Gonghong, Y., Jiang, Y., & Nickels, J. T. (2013). Cell Cycle PP2A Cdc55 652 regulates G 1 cyclin stability. Cell Cycle 1201 Cell Cycle, 12, 1201–1210. 653 https://doi.org/10.4161/cc.24231 654 Mendenhall, M. D., & Hodge, A. E. (1998). Regulation of Cdc28 cyclin-dependent protein kinase activity 655 during the cell cycle of the yeast Saccharomyces cerevisiae. Microbiology and Molecular Biology 656 Reviews : MMBR, 62(4), 1191–1243. http://www.ncbi.nlm.nih.gov/pubmed/9841670 657 Meyers, A., del Rio, Z. P., Beaver, R. A., Morris, R. M., Weiskittel, T. M., Alshibli, A. K., Mannik, J., Morrell-658 Falvey, J., & Dalhaimer, P. (2016). Lipid Droplets Form from Distinct Regions of the Cell in the 659 Fission Yeast Schizosaccharomyces pombe. Traffic, 17(6), 657–669. 660 https://doi.org/10.1111/TRA.12394 661 Monteiro, F., Hubmann, G., Takhaveev, V., Vedelaar, S. R., Norder, J., Hekelaar, J., Saldida, J., Litsios, A., 662 Wijma, H. J., Schmidt, A., & Heinemann, M. (2019). Measuring glycolytic flux in single yeast cells 663 with an orthogonal synthetic biosensor. Molecular Systems Biology, 15(12), e9071. 664 https://doi.org/10.15252/MSB.20199071 665 Moreno-Torres, M., Jaquenoud, M., & De Virgilio, C. (2015). TORC1 controls G1–S cell cycle transition in 666 yeast via Mpk1 and the greatwall kinase pathway. Nature Communications 2015 6:1, 6(1), 1–10. 667 https://doi.org/10.1038/ncomms9256 668 Müller, D., Exler, S., Aguilera-Vázquez, L., Guerrero-Martín, E., & Reuss, M. (2003). Cyclic AMP mediates 669 the cell cycle dynamics of energy metabolism in Saccharomyces cerevisiae. Yeast (Chichester, 670 England), 20(4), 351–367. https://doi.org/10.1002/YEA.967 671 Nickels, J. T., & Broach, J. R. (1996). A ceramide-activated protein phosphatase mediates ceramide-672 induced G1 arrest of Saccharomyces cerevisiae. Genes & Development, 10(4), 382–394. 673 https://doi.org/10.1101/GAD.10.4.382 674 Novarina, D., Koutsoumpa, A., & Milias-Argeitis, A. (2022). A user-friendly and streamlined protocol for 675 CRISPR/Cas9 genome editing in budding yeast. STAR Protocols, 3(2), 101358. 676 https://doi.org/10.1016/J.XPRO.2022.101358 677 Oelkers, P., Cromley, D., Padamsee, M., Billheimer, J. T., & Sturley, S. L. (2002). The DGA1 gene 678 determines a second triglyceride synthetic pathway in yeast. The Journal of Biological Chemistry, 679 277(11), 8877–8881. https://doi.org/10.1074/JBC.M111646200 680 Oelkers, P., Tinkelenberg, A., Erdeniz, N., Cromley, D., Billheimer, J. T., & Sturley, S. L. (2000). A Lecithin 681 Cholesterol Acyltransferase-like Gene Mediates Diacylglycerol Esterification in Yeast*. Journal of 682 Biological Chemistry, 275(21), 15609–15612. https://doi.org/10.1074/JBC.C000144200 683 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 20 Ouahoud, S., Fiet, M. D., Martínez-Montañés, F., Ejsing, C. S., Kuss, O., Roden, M., & Markgraf, D. F. 684 (2018). Lipid droplet consumption is functionally coupled to vacuole homeostasis independent of 685 lipophagy. Journal of Cell Science, 131(11). https://doi.org/10.1242/JCS.213876/265618/AM/LIPID-686 DROPLET-CONSUMPTION-IS-FUNCTIONALLY-COUPLED 687 Papagiannakis, A., Niebel, B., Wit, E. C., & Heinemann, M. (2017). Autonomous Metabolic Oscillations 688 Robustly Gate the Early and Late Cell Cycle. Molecular Cell, 65(2), 285–295. 689 https://doi.org/10.1016/j.molcel.2016.11.018 690 Pedregosa, F., Michel, V., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Vanderplas, J., Cournapeau, 691 D., Varoquaux, G., Gramfort, A., Thirion, B., Dubourg, V., Passos, A., Brucher, M., Perrot, M., & 692 Duchesnay, É. (2011). Scikit-learn: Machine Learning in Python. Journal of Machine Learning 693 Research, 12, 2825–2830. http://scikit-learn.sourceforge.net. 694 Qiu, F., Kang, N., Tan, J., Yan, S., Lin, L., Cai, L., Goodman, J. M., & Gao, Q. (2023). Fatty Acyl Coenzyme A 695 Synthetase Fat1p Regulates Vacuolar Structure and Stationary-Phase Lipophagy in Saccharomyces 696 cerevisiae. Microbiology Spectrum, 11(1). https://doi.org/10.1128/SPECTRUM.04625-697 22/SUPPL_FILE/SPECTRUM.04625-22-S0005.MP4 698 Rajakumari, S., & Daum, G. (2010). Janus-faced enzymes yeast Tgl3p and Tgl5p catalyze lipase and 699 acyltransferase reactions. Molecular Biology of the Cell, 21(4), 501–510. 700 https://doi.org/10.1091/MBC.E09-09-0775/ASSET/IMAGES/LARGE/ZMK0041093470009.JPEG 701 Santos-Rosa, H., Leung, J., Grimsey, N., Peak-Chew, S., & Siniossoglou, S. (2005). The yeast lipin Smp2 702 couples phospholipid biosynthesis to nuclear membrane growth. The EMBO Journal, 24(11), 1931–703 1941. https://doi.org/10.1038/SJ.EMBOJ.7600672 704 SGD. (2017). Saccharomyces Genome Database. http://yeastgenome.org/ 705 Silljé, H. H. W., Ter Schure, E. G., Rommens, A. J. M., Huls, P. G., Woldringh, C. L., Verkleij, A. J., Boonstra, 706 J., & Verrips, C. T. (1997). Effects of different carbon fluxes on G1 phase duration, cyclin expression, 707 and reserve carbohydrate metabolism in Saccharomyces cerevisiae. Journal of Bacteriology, 708 179(21), 6560–6565. https://doi.org/10.1128/JB.179.21.6560-6565.1997 709 Su, W. M., Han, G. S., & Carman, G. M. (2014). Cross-talk Phosphorylations by Protein Kinase C and 710 Pho85p-Pho80p Protein Kinase Regulate Pah1p Phosphatidate Phosphatase Abundance in 711 Saccharomyces cerevisiae*. Journal of Biological Chemistry, 289(27), 18818–18830. 712 https://doi.org/10.1074/JBC.M114.581462 713 Su, W. M., Han, G. S., Dey, P., & Carman, G. M. (2018). Protein kinase A phosphorylates the Nem1–Spo7 714 protein phosphatase complex that regulates the phosphorylation state of the phosphatidate 715 phosphatase Pah1 in yeast. Journal of Biological Chemistry, 293(41), 15801–15814. 716 https://doi.org/10.1074/JBC.RA118.005348 717 Takhaveev, V., Özsezen, S., Smith, E. N., Zylstra, A., Chaillet, M. L., Chen, H., Papagiannakis, A., Milias-718 Argeitis, A., & Heinemann, M. (2023). Temporal segregation of biosynthetic processes is 719 responsible for metabolic oscillations during the budding yeast cell cycle. Nature Metabolism 2023 720 5:2, 5(2), 294–313. https://doi.org/10.1038/s42255-023-00741-x 721 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 21 Terpstra, H. M., Gómez-Sánchez, R., Veldsink, A. C., Otto, T. A., Veenhoff, L. M., & Heinemann, M. 722 (2024). PunctaFinder: An algorithm for automated spot detection in fluorescence microscopy 723 images. Https://Doi.Org/10.1091/Mbc.E24-06-0254, 35(12). https://doi.org/10.1091/MBC.E24-06-724 0254 725 Thévenaz, P., Ruttimann, U. E., Unser, M., & Member, S. (1998). A Pyramid Approach to Subpixel 726 Registration Based on Intensity. In IEEE TRANSACTIONS ON IMAGE PROCESSING (Vol. 7, Issue 1). 727 Ueno, K., Nagano, M., Shimizu, S., Toshima, J. Y., & Toshima, J. (2016). Lipid droplet proteins, Lds1p, 728 Lds2p, and Rrt8p, are implicated in membrane protein transport associated with ergosterol. 729 Biochemical and Biophysical Research Communications, 475(4), 315–321. 730 https://doi.org/10.1016/j.bbrc.2016.05.099 731 Verduyn, C., Postma, E., Scheffers, W. A., & Van Dijken, J. P. (1992). Effect of benzoic acid on metabolic 732 fluxes in yeasts: A continuous-culture study on the regulation of respiration and alcoholic 733 fermentation. Yeast, 8(7), 501–517. https://doi.org/10.1002/yea.320080703 734 Wang, C. W., Chen, R. H., & Chen, Y. K. (2024). The lipid droplet assembly complex consists of seipin and 735 four accessory factors in budding yeast. Journal of Biological Chemistry, 300(8). 736 https://doi.org/10.1016/j.jbc.2024.107534 737 Wang, C. W., Miao, Y. H., & Chang, Y. S. (2014). A sterol-enriched vacuolar microdomain mediates 738 stationary phase lipophagy in budding yeast. Journal of Cell Biology, 206(3), 357–366. 739 https://doi.org/10.1083/JCB.201404115/VIDEO-1 740 Weill, U., Yofe, I., Sass, E., Stynen, B., Davidi, D., Natarajan, J., Ben-Menachem, R., Avihou, Z., Goldman, 741 O., Harpaz, N., Chuartzman, S., Kniazev, K., Knoblach, B., Laborenz, J., Boos, F., Kowarzyk, J., Ben-742 Dor, S., Zalckvar, E., Herrmann, J. M., … Schuldiner, M. (2018). Genome-wide SWAp-Tag yeast 743 libraries for proteome exploration. Nature Methods, 15(8), 617–622. 744 https://doi.org/10.1038/s41592-018-0044-9 745 Wilfling, F., Wang, H., Haas, J. T., Krahmer, N., Gould, T. J., Uchida, A., Cheng, J. X., Graham, M., 746 Christiano, R., Fröhlich, F., Liu, X., Buhman, K. K., Coleman, R. A., Bewersdorf, J., Farese, R. V., & 747 Walther, T. C. (2013). Triacylglycerol Synthesis Enzymes Mediate Lipid Droplet Growth by 748 Relocalizing from the ER to Lipid Droplets. Developmental Cell, 24(4), 384. 749 https://doi.org/10.1016/J.DEVCEL.2013.01.013 750 Wunderling, K., Zurkovic, J., Zink, F., Kuerschner, L., & Thiele, C. (2023). Triglyceride cycling enables 751 modification of stored fatty acids. Nature Metabolism 2023 5:4, 5(4), 699–709. 752 https://doi.org/10.1038/s42255-023-00769-z 753 Yang, P. L., Hsu, T. H., Wang, C. W., & Chen, R. H. (2016). Lipid droplets maintain lipid homeostasis 754 during anaphase for efficient cell separation in budding yeast. Molecular Biology of the Cell, 27(15), 755 2368–2380. https://doi.org/10.1091/MBC.E16-02-0106 756 Yang, S., Wang, Y., Huang, S., Zhang, T., Xu, P., Jiang, C., & Ye, C. (2025). Temporal oscillation of 757 phospholipids promotes metabolic efficiency. Nature Chemical Biology, 1–12. 758 https://doi.org/10.1038/S41589-025-01885-759 5;SUBJMETA=1194,287,45,60,631,92;KWRD=BIOSYNTHESIS,PHOSPHOLIPIDS 760 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 22 Zhao, G., Chen, Y., Carey, L., & Futcher, B. (2016). Cyclin-Dependent Kinase Co-Ordinates Carbohydrate 761 Metabolism and Cell Cycle in S. cerevisiae. Molecular Cell, 62(4), 546–557. 762 https://doi.org/10.1016/J.MOLCEL.2016.04.026 763 764 765 766 767 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 23 SUPPLEMENTARY INFORMATION 768 769 Figure S1. BODIPY-TR fluorescence is not detected in the GFP channel and puncta of GFP 770 autofluorescence are not detected in cells expressing an LD reporter tagged with mNeonGreen. (A) 771 Colocalisation between LDs stained with BODIPY-TR and puncta detected in the GFP channel in wild-type 772 cells. The left circle of the Venn diagram represents all LDs identified with BODIPY-TR in the RFP channel; 773 the right circle represents all puncta identified in the GFP channel. The orange overlapping region are 774 puncta that colocalise between the LDs stained with BODIPY-TR and the puncta in the GFP channel. The 775 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 24 red region indicates LDs that do not colocalise with a punctum in the GFP channel, while the green 776 region indicates puncta in the GFP channel that do not colocalise with an LD stained with BODIPY-TR. 777 The white regions represent puncta with ambiguous colocalisation: colocalisation between puncta in the 778 two fluorescence channels is only probable, not certain, despite their midpoints being close together. 779 274 cells were analysed, 427 BODIPY-TR puncta and 89 puncta in the GFP channel were identified; (B) 780 Comparison of the average fluorescence intensity measured in the GFP channel for cells stained with 781 BODIPY-TR and unstained cells. Since the cells stained with BODIPY-TR are not brighter than the 782 unstained cells, we can conclude that BODIPY-TR fluorescence is not detected in the GFP channel ; (C) 783 Fluorescence microscopy images of wild-type cells, recorded in the GFP channel . The top row shows 784 unstained cells, while the bottom row shows cells stained with the red fluorophore BODIPY-TR . There 785 are no clearly visible differences between stained and unstained cells, demonstrating that BODIPY-TR 786 staining does not influence images recorded in the GFP channel. Furthermore, puncta could be detected 787 in images of both stained and unstained cells, as shown in the third column of images. This finding 788 indicates that brighter spots in the autofluorescence occur naturally and can result in the appearance of 789 puncta; (D) Fluorescence microscopy images and detected puncta of three wild-type cells stained with 790 BODIPY-TR. The puncta identified in the GFP channel images are only slightly brighter than the 791 cytoplasm and, in the visualised cells as well as the majority of other cells, do not colocalise with 792 BODIPY-TR puncta, ruling out that these GFP puncta are due to detection of BODIPY-TR signal in the GFP 793 channel; (E-F) Average fluorescence intensity of the cytoplasm of cells with at least one punctum, the 794 same whole cells, i.e. cytoplasm and puncta combined, and the detected puncta. Shaded areas in E and 795 F indicate the box (first quartile to third quartile) of whole wild-type cells or puncta detected in wild-796 type cells, respectively. The punctum detection threshold values to detect GFP puncta in wild-type cells 797 are more stringent than those applied to cells expressing mNG-Tgl3 or Rrt8-mNG (Table S4) . Still, the 798 average fluorescence intensity of the cytoplasm is significantly lower than that of whole cells in the 799 three reporter strains, but not in the wild-type control (one-sided Mann-Whitney U-test, p<0.01). This 800 finding indicates that the GFP autofluorescence puncta detected in the wild type are similar to the 801 cytoplasm with regards to fluorescence intensity, while in the three LD reporter strains, the bright 802 puncta cause the average fluorescence intensity of whole cells to be higher than that of the cytoplasmic 803 region alone. Moreover, the average fluorescence intensity of the GFP autofluorescence puncta 804 detected in the wild type is notably lower than that of puncta detected in any of the three LD reporter 805 strains. Together, these results indicate that it is improbable that GFP autofluorescence puncta are 806 detected in the three LD reporter strains. 807 808 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 25 809 Figure S2. Cell cycle length increases with experiment duration in time-lapse imaging of Rrt8-mNG but 810 not Pln1-mNG and mNG-Tgl3. (A) Cell cycle length, from one instance of mitotic exit (ME) to the next 811 was plotted against the starting time of the cycle within the time-lapse experiment for cells expressing 812 Pln1-mNG, mNG-Tgl3 or Rrt8-mNG as a reporter protein for LDs. Replicate experiments are shown with 813 distinct colours . Trend lines describing cell cycle length as a function of cycle initiation time were 814 obtained with linear regression, which was performed on pooled data of replicate experiments; (B) 815 Bootstrapping was performed to estimate the variation in the slope of the regression lines from A. For 816 each bootstrap iteration, 50% of the data was randomly sampled with replacement and linear regression 817 was performed to obtain a slope value; a total of 100 iterations were performed for each genetic 818 background. Grey horizontal lines indicate the slope values of the regression lines in A, which were 819 obtained using all data points. Mean and standard deviation of the slopes obtained with bootstrapping 820 are indicated in black. Interestingly, regression lines fitted to data from cells expressing Rrt8-mNG ha ve 821 positive slope of values, while regression lines fitted to data from cells expressing Pln1-mNG or mNG-822 Tgl3 have an average slope value of approximately 0. Thus, in time-lapse imaging of cells expressing 823 Rrt8-mNG cell cycle length increases the longer the experiment has lasted. This reveals a potential 824 phototoxic effect, caused by the cumulative effects of repeated exposure to the lasers used for 825 fluorophore excitation. 826 827 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 26 828 Figure S3. LD number is dynamic along the cell cycle while LD size is constant. LDs were identified in 829 time-lapse microscopy images of cells expressing either Pln1-mNG (A, C, E) or mNG-Tgl3 (B, D, F) as an 830 LD marker protein. For both reporter strains, three replicate experiments were performed. Cell cycle 831 trajectories were aligned from one occurrence of mitotic exit (ME) to the next (red vertical lines at cell 832 cycle progression values 0 and 1) and were also aligned for START (bright green vertical line) and 833 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 27 budding (orange vertical line). (A-D) Density estimations, showing densely populated data points in red 834 and sparsely populated data points in blue, obtained with Gaussian kernel estimation , show the cell 835 cycle dynamics of (A , B) the number of LDs normalised to the cell cross-area and (C , D) LD size. The 836 number of cell cycles assessed in each replicate is indicated on the right side of each plot. The plots in 837 each second column zoom in on the plots with all data and show their most densely populated region, 838 located between the dashed horizontal lines . Here, t he colour map has been rescaled to assess the 839 density in more detail. Notably, the density plots show the same cell cycle dynamics of the LD number 840 normalised to the cell cross-area predicted with Gaussian process regression (Figure 2A) when Pln1-841 mNG is used as an LD reporter protein, but not with mNG-Tgl3. Still, with mNG-Tgl3 as an LD marker, 842 relatively dense subpopulations with <0.1 LDs/µm 2 are visible early in the cell cycle in all three 843 replicates, reflecting the trough around START in the cell cycle dynamics of the number of LDs per cell 844 cross-area predicted with Gaussian process regression (Figure 2 B). Also, with mNG- Tgl3 as an LD 845 reporter protein, distinct subpopulations of cells with one or two puncta are visible, at #LD/area values 846 of approximately 0.07 LDs/µm -2 and 0.14 LDs/µm -2, respectively. These subpopulations occur since the 847 area-normalised LD number is obtained by dividing the discrete number of LDs by the continuous cell 848 cross-area. Evidently, the range in cell cross-area values is narrow, causing the resulting area-normalised 849 LD number to still appear discrete in the density plots; (E-F) Gaussian process regression outputs 850 showing the cell cycle dynamics of the total area of detected LDs , i.e. summed sizes of all LDs detected 851 in a cell, without normalisation to the cell cross-area. Since these outputs resemble the cell cycle 852 dynamics of the area-normalised LD number, its oscillation does not result from the normalisation to the 853 cell cross-area. Moreover, the strong resemblance between the cell cycle dynamics of the summed LD 854 size and the dynamics of the area-normalised LD number further confirms that LD number, but not LD 855 size, is dynamic along the cell cycle. 856 857 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 28 858 Figure S4. Genetic perturbation of TAG metabolism affects Pln1-mNG expression levels. TAG metabolism 859 was perturbed by gene deletion of either DGA1 and LRO1, which encode the main TAG synthases, or 860 TGL3 and TGL4, which encode the lipases responsible for TAG mobilisation from LDs. Average Pln1-861 mNeonGreen fluorescence intensity was determined in snapshot images of cells from an exponential 862 culture for (A) the whole cell and the cytoplasmic region as well as (B) the LDs in the wild type, 863 ΔDGA1ΔLRO1 and ΔTGL3ΔTGL4. White diamonds indicate the median and open circles indicate the 864 population average. Both genetic perturbations led to significant changes in measured Pln1-865 mNeonGreen fluorescence compared to the wild type (two-sided Mann-Whitney U-test, p < 0.001) for 866 all three regions of interest. Notably, the change in Pln1-mNG fluorescence compared to the wild type is 867 much larger for ΔDGA1ΔLRO1 than for ΔTGL3ΔTGL4, as quantified using Cohen’s d to assess the effect 868 size. For ΔDGA1ΔLRO1, the effect size for the changing fluorescence intensity between deletion strain 869 and wild type was 2.54 , 2.40 and 1.86 for the whole cell, the cytoplasm and the puncta, respectively. In 870 contrast, for ΔTGL3ΔTGL4, the effect size for these three regions of interest was equal to 0.54, 0.49 and 871 0.79. 872 873 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 29 874 Figure S5 . Duration of the cell cycle and its subphases are altered slightly in ΔDGA1ΔLRO1 and 875 ΔTGL3ΔTGL4 compared to the wild type. Probability density functions for duration of (A) the whole cell 876 cycle and the cell cycle phases (B) mitotic exit to START, (C) START to budding and (D) budding to mitotic 877 exit in the wild type, ΔDGA1ΔLRO1 and ΔTGL3ΔTGL4. To obtain these distributions, cell cycles recorded 878 in three replicate experiments with the wild type and two replicate experiments each with ΔDGA1ΔLRO1 879 and ΔTGL3ΔTGL4 were pooled. Solid and dotted vertical lines denote the mean and median value of 880 each distribution, respectively . Percentages denote the change in median and mean in each deletion 881 strain compared to the wild type. Above each plot, a schematic representation of the cell cycle indicates 882 the cell cycle phase(s) studied. Notably, only cell cycles with a total duration of at most 180 min were 883 included in the analysis. 884 885 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 30 Table S1. Yeast strains used in the current study. 886 Strain Source YSBN6 (S288C-derived strain, MATa FY3 HO::HphMX4) Canelas et al., 2010 YSBN6 WHI5::mCherry-BLE Litsios et al., 2021 YSBN6 PLN1::mNeonGreen This study YSBN6, TGL3::mNeonGreen-Tgl3 This study YSBN6 RRT8::mNeonGreen This study YSBN6 WHI5::mCherry-BLE PLN1::mNeonGreen This study YSBN6 WHI5::mCherry-BLE TGL3::mNeonGreen-Tgl3 This study YSBN6 WHI5::mCherry-BLE RRT8:mNeonGreen This study YSBN6 WHI5::mCherry-BLE PLN1::mNeonGreen ΔDga1 ΔLro1 This study YSBN6 WHI5::mCherry-BLE PLN1::mNeonGreen ΔTgl3 ΔTgl4 This study 887 Table S2. Primers used in the current study. 888 Name Sequence Application PLN1_sg_fwd gactttCTAATTGGTCGACACAGCCG Primers with sgRNA guide sequences targeting the specified genes within the S. cerevisiae genome. Upper-case nucleotides encompass the actual guide sequences, while lower-case nucleotides comprise adapters that allow plasmid assembly in a GoldenGate Assembly approach. PLN1_sg_rev aaacCGGCTGTGTCGACCAATTAGa a TGL3_sg_fwd gactttTGAGTTGCCGTTAAGCATGA TGL3_sg_rev aaacTCATGCTTAACGGCAACTCAaa RRT8_sg_fwd gactttTGGTGTACTTCGCTACTAAA RRT8_sg_rev aaacTTTAGTAGCGAAGTACACCAa a TGL4_sg_fwd gactttTTTACTCAATAAGAAAACAC TGL4_sg_rev aaacGTGTTTTCTTATTGAGTAAAaa DGA1_sg_fwd gactttTTGGGTAATAATGAATTCAT DGA1_sg_rev aaacATGAATTCATTATTACCCAAaa LRO1_sg_fwd gactttGATGGATAGTGAGTCAATGT LRO1_sg_rev aaacACATTGACTCACTATCCATCaa PLN1_repair_fwd TGGGCAATGCCACCATTGAGAAGC TAAAGGCCTCAAGAGAAGACCAAA CCAATTCTAAGCCAGCGGCTGTGT CGACCAATATGGTGAGCAAGGGC GAG Primers to create repair fragments that introduce an mNeonGreen tag on the specified target proteins with CRISPR-Cas9 assisted cutting. Template in the PCR is the DNA sequence encoding mNeonGreen, codon-optimised for S. cerevisiae. PLN1_repair_rev TAACTATATAAGAGTGGCAGGAAA AAAAATCAGGCGCACGATTAGCGC AAAACCAAATTATTACTTGTACAGC TCGTCCATGC TGL3_tag_repair_fwd ATGACACAATAGTAAGGGAATCAT CTATTCATATATCACATCTTTGAGTT GCCGTTAAGCATGGTGAGCAAGG GCGAG TGL3_tag_repair_rev GTATCCAGTTTTTCAAAAGGGTCG GTATTACAGCAGACACCTTGTATTC .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 31 CTGCGCCGTTTCCTTCATCTTGTAC AGCTCGTCCATGCCC RRT8_repair_fwd AATAGTCACCATATCTAGCAACACT GTTGGTGCAGCTAAATGGTGTACT TCGCTACTGAAGGGTGAAAGAAAG AAGGAAATGGTGAGCAAGGGCGA G RRT8_repair_rev GATTAACAATTAGTTAAGGAATAT ATAATCACACTTCTACATAAATTTG CTGTTTTAGGCTTACTTGTACAGCT CGTCCATGC PLN1_rep_short_fwd GGTTGGACTTGGGCAATG Primers to amplify repair fragments that introduce an mNeonGreen tag on the specified target protein. Template in the PCR is genomic DNA from a strain that already expresses the tagged target protein. PLN1_rep_short_rev CAAATAACTATATAAGAGTGGCAG G TGL3_rep_short_fwd ATGACACAATAGTAAGGGAATCAT C TGL3_rep_short_rev CATACACTACACGCAGTATCCAG RRT8_rep_short_fwd AATAGTCACCATATCTAGCAAC RRT8_rep_short_rev GAACTTGATTAACAATTAGTTAAG G TGL3_deletion_repair_fwd AGTAAGGGAATCATCTATTCATATA TCACATCTTTGAGTTGCCGTTAAGC tatcgtttccacttttttctgtc Primers to amplify repair fragments that delete the gene of interest from the S. cerevisiae genome after CRISPR-Cas9 assisted cutting. Nucleotides shown in lower case are reverse complementary to the other primer for amplification in a template-free PCR. TGL3_deletion_repair_rev ATCGAGCTCTATCAATAAAAAAAA TAAGACAGAAAAAAGTGGAAACG ATAgcttaacggcaactcaaagatg TGL4_repair_fwd CGCTGTAATAATTATTGAAGGGAG TACAGGTATATGTAATAAAAGTCT GAgaaaacacgggcttg TGL4_repair_rev GGCCATTCGAATAAATACATAGAT GAAAAAGAATATCTAGAGGATATA TAAGCAAGCCCGTGTTTTCtcagactt ttattacatatacctg DGA1_repair_fwd TACATATACATAAGGAAACGCAGA GGCATACAGTTTGAACAGTCACAT AAtaataatgaattcattggaaaac DGA1_repair_rev CTTAAGATATACAGCCCAAACACTA AAAAATCCTTATTTATTCTAACATA TTTTGTGTTTTCCAATGAATTCATTA TTA LRO1_repair_fwd CCATTACAAAAGGTTCTCTACCAAC GAATTCGGCGACAATCGAGTAAAA Ataaatgaccgacattgactcactatc LRO1_repair_rev GCGACGCGCCTTCTTTTCGCTCTTT GAAATAATACACGGATGGATAGTG AGTCAATGTCGGTCATTTA PLN1_check_fwd CGAAACCTACCAACGCTTCAC Primer pairs to verify that the .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 32 PLN1_check_rev GTCTCTTGATCGAGCTATAACC desired genomic modifications were successful. TGL3_tag_check_fwd CCTAGGTCTGAAAATTCAACCC TGL3_tag_check_rev ATGACTCTTGAGTGTGGCCG RRT8_check_fwd CATATGTTTCGGTATGTCTGCC RRT8_check_rev GACGAGCAAGTTTTATCGAACG TGL3_deletion_check_fwd AGATACTTATCCTAGGTCTG TGL3_deletion_check_rev CTGAATGAGAAGGAGTCAAC TGL4_check_fwd TAATTGCGACTATGAAACGC TGL4_check_rev ACCAATATCTTTCTTCCACC DGA1_check_fwd CTTTCACTACACTTCCGCCAAAG DGA1_check_rev CCTAAACTTACATTCAAACAACTTC LRO1_check_fwd CCAACTACTTAGTGTAGATC LRO1_check_rev CTCCTCTATCTACTGTCGTTTG mNG_seq_rev CCATCATTAGGGTTACCTG Sequencing primers to verify that the DNA sequence encoding mNeonGreen has been integrated at the target site correctly. The reverse primer anneals approximately 130 bp from the beginning of mNeonGreen, while the forward primer anneals approximately 170 bp before its end. mNG_seq_fwd GCTAGAACAACGTACACATTCG 889 Table S3. Imaging settings used for microscopy experiments. For each fluorophore and protein of 890 interest, the imaging channel, light intensity and excitation time applied are detailed below. 891 Target Imaging channel Light intensity Excitation time BODIPY-TR (fixed cells) RFP 1% 5 ms Pln1-mNG (fixed cells) GFP 7% 300 ms mNG-Tgl3 (fixed cells) GFP 7% 300 ms Rrt8-mNG (fixed cells) GFP 7% 300 ms autofluorescence (fixed cells) GFP 7% 300 ms Whi5-mCherry (live cells; time-lapse) RFP 10% 300 ms Pln1-mNG (live cells; time-lapse) GFP 3% 200 ms mNG-Tgl3 (live cells; time-lapse) GFP 3% 200 ms 892 Table S4. Threshold values used for automated punctum detection with PunctaFinder . Threshold values 893 for punctum detection in fluorescence microscopy images of the neutral lipid dye BODIPY-TR (RFP 894 channel) and mNeonGreen (mNG) tagged reporter proteins for LDs or a wild-type autofluorescence 895 control (GFP channel). Because of slight differences in BODIPY-TR staining between the four samples, 896 separate thresholds were determined for punctum detection in the RFP channel as well. In all cases, the 897 punctum diameter was set to three pixels and the overlap parameter value to zero. Thresholds were 898 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint 33 determined based on manually validated datasets of 42 cells expressing Pln1-mNG, 41 cells expressing 899 mNG-Tgl3, 43 cells expressing Rrt8-mNG and 45 wild-type cells. 900 Genetic

Fluorescence channel T ratio, local T ratio, global T CV Pln1-mNG GFP 1.408 1.406 0.358 RFP 1.252 1.338 0.262 mNG-Tgl3 GFP 1.342 1.508 0.272 RFP 1.264 1.364 0.252 Rrt8-mNG GFP 1.348 1.353 0.242 RFP 1.254 1.398 0.206 wild type GFP 1.380 1.648 0.266 RFP 1.222 1.382 0.194 901 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted September 25, 2025. ; https://doi.org/10.1101/2025.09.25.678471doi: bioRxiv preprint