We recently developed a synthetic protein based system that forms condensates in vivo (Heidenreich et al. 2020) . The system consists of two proteins that interact and condense into a round, mesh-like structure. The structures grow over time, since the components are constitutively expressed. This hints that condensate size could be used as a molecular timer. Here, we show that such a system indeed serves as a good proxy for yeast's replicative age.

The previously developed version of the system was designed to yield heterogeneous expression of the two components in order to map phase diagrams. With such heterogeneous expression from plasmids, a majority of cells contain an excess of one component and do not form any condensate. Here however, our goal is that all cells contain a reporter of their replicative age, which requires the two components to be expressed in stoichiometric amounts. Therefore, we integrated the two constructs into two alleles of the same locus of diploid cells. Furthermore, to decrease the stoichiometry dependence of the formation of condensates, we increased the valency of one component from four to ten. Indeed, it was shown that increasing valency enhances phase separation of two-component systems over a wider concentration range (Nandi et al. 2019; Banani et al. 2016) . Finally, to assess the effect of affinity on the age-reporting reliability of our system, we utilized two different affinities between the interaction domains of the two components (Figure 1a).

As expected, condensates are only formed when both components are present, after mating two haploid strains where each haploid carries one component (Figure 1b). Timelapses of budscar-stained cells expressing our synthetic system revealed that condensates become detectable within the first cell-cycle of a newly born cell (Figure 1c, Movie 1).

In a dividing cell population, the frequency f _A of cells of age A is f _A = 1 / 2 ^A , so that half of the population has not yet divided (age 1), a quarter of the population is one cycle old (age 2) and so on (Figure 1d). To examine whether the condensate’s size recapitulates such an age-distribution, we measured the condensate sizes across a large number of cells (N), binned the results into log ₂ (N) bins, and compared it to the theoretically expected distribution of ages. We observed that the size-distribution of condensates in the population matched the expected distribution of ages, and the higher affinity version of the system appeared more accurate (Figure 1e). To further validate that condensate size serves as a good proxy for cell age, we stained and manually counted budscars of cells expressing our synthetic system, and correlated this number to the size of automatically identified condensates (Figure 1f). The number of budscars, and therefore yeast replicative age, correlated well with condensate size (R ² =0.83 and 0.87 for the low- and high affinity versions of our system, respectively, Figure 1g). Interestingly, such correlation implies that these synthetic condensates are rarely passed to daughter cells, presumably due to their size prohibiting passage through the bud neck.

The identification or isolation of old cells by micro-dissection is laborious. Alternative methods, such as microfluidic techniques (Jo et al. 2015; Lee et al. 2012; Chen, Crane, and Kaeberlein 2017; Crane et al. 2014) , the mother enrichment program (Lindstrom and Gottschling 2009) , or a magnetic-bead based enrichment of aged cells (Hendrickson et al. 2018) are difficult to upscale for genome-wide studies. The approach presented here, where a protein’s assembly serves as a reporter for age, could be a powerful alternative owing to the ease with which age can be inferred by this method. More generally, synthetic assemblies in cells find unexpected applications, such as the use of synthetic protein filaments (Garcia-Seisdedos et al. 2017) as ticker-tape recorders of intra-cellular events (Linghu et al. 2021; Lin et al. 2021) . It is our hope that this system will serve the community in novel applications.

The components were adapted from our previously developed system (Heidenreich et al. 2020) . We used the dimer fused to FusionRed and Im2 with either wild type (1.5 ± 0.1 x 10 ^-8 M), or low affinity (E41A) for E9: 3*10 ^-5 M (Li et al. 1998) , in combination with a decameric oligomerization domain (1VPX (Joint Center for Structural Genomics (JCSG) 2004) ) fused to E9 and Venus. To generate homogeneous expression, both components were driven by the same promoter (TDH3 promoter) and terminator (CYC terminator). The constructs were cloned into m3925 plasmids (Voth, Jiang, and Stillman 2003) that were modified to carry G418 (decamer) or hygromycin (dimer) resistance. Finally, the constructs were inserted into the TRP1 locus of of yeast strains of opposite mating type, where the dimer was inserted into MATα type (BY4741), and the decamer into MATa type (BY4742) cells (Brachmann et al. 1998) . The two resulting strains were mated, yielding diploid cells (BY4743) carrying both components (Figure 1 a and b). The plasmid and strain details are listed in Table 1 (Reagents).

Cells were grown in synthetic defined (SD) media with hygromycin and G418 to logarithmic growth (OD ₆₀₀ 0.6-0.8) overnight. For staining budscars, cells were transferred to optical 96-well plates (Greiner ^TM ) and WGA-CF405M (Biotium) was added to a final concentration of 50 μg/ml 20-30 min prior to imaging. Images were acquired using an Olympus IX83 microscope with a Yokogawa CSU-W1 spinning disc confocal scanner and an automated piezo-stage (Mad City Labs). For the time-lapse movie of budscar stained cells (Figure 1c), seven z-stacks were acquired and maximum intensity z-projections were generated using FIJI (Schindelin et al. 2012) . For analyzing the condensate’s integrated intensity, 100 images with eight z-stacks were acquired and an average z-projection was used (Figure 1e-h). Cells and condensates were segmented, and the condensate’s intensity was recorded using custom scripts (Matalon et al. 2018) . To correlate the condensate’s integrated intensity to the number of budscars, cells and condensates were segmented automatically. Then, tens of cells with each number of budscars were identified manually and the preceding segmentation was used to measure the corresponding condensate’s integrated intensity. Data analysis and plotting was conducted using custom R scripts.

Table 1: Plasmids and strains used and generated in this study.