Eliciting the impacts of cellular noise on metabolic trade-offs by quantitative mass imaging

Abstract

Optimal metabolic trade-offs between growth and productivity are key constraints in strain optimization by metabolic engineering; however, how cellular noise impacts these trade-offs and drives the emergence of subpopulations with distinct resource allocation strategies, remains largely unknown. Here, we introduce a single-cell strategy for quantifying the trade-offs between triacylglycerol production and growth in the oleaginous microorganism Yarrowia lipolytica. The strategy relies on high-throughput quantitative-phase imaging and, enabled by nanoscale secondary ion mass spectrometry analyses and dedicated image processing, allows us to image how resources are partitioned between growth and productivity. Enhanced precision over population-averaging biotechnologies and conventional microscopy demonstrates how cellular noise impacts growth and productivity differently. As such, subpopulations with distinct metabolic trade-offs emerge, with notable impacts on strain performance and robustness. By quantifying the self-degradation of cytosolic macromolecules under nutrient-limiting conditions, we discover the cell-to-cell heterogeneity in protein and fatty-acid recycling, unmasking a potential bet-hedging strategy under starvation.

Fig. 1

Metabolic trade-offs by quantitative mass imaging. a Schematic illustrating substrate uptake and resource partitioning to growth and production, as well as the underlying trade-offs between these two metabolic objectives. b Quantitative phase-imaging (QPI) enables the independent localization (x,y,z) and phase-delay quantification of the cell cytosol (ΔΦ_cytosol) and TAG loaded lipid droplets (ΔΦ_TAG). ΔΦ_cytosol and ΔΦ_TAG are subsequently converted to their corresponding dry-mass values, enabling trade-off phenotyping between growth and TAG production

Fig. 2

Quantitative mass imaging and cell-to-cell lipid-content heterogeneity. a An optical-phase image of individual Y. lipolytica cells labeled from (i) to (iv); arrows indicate the cytosolic LDs, and scale-bar is displayed in radians. b Histogram of the lipid-content in % volume (V_TAG/V_biomass) and dry-mass (DM_TAG/DM_biomass) ratios for the cells shown in a; importantly, the single-cell volumetric lipid-content is seen to scale inversely with the DM lipid-content specifically for cells (i), (ii), and (iii)

Fig. 3

Elemental composition of Y. lipolytica. a NanoSIMS images of the MTYL038 strain at two C/N growth conditions for 6, 15, and 100 h (Methods); the cytosolic pools of naturally abundant ¹⁴N and the LD content of ¹³C are highlighted in red and green, respectively. b Box-plots of ¹²C¹⁴N/¹²C¹³C ratio of the cytosolic LDs droplets (red), the cytosol excluding the LDs (blue), and the extracellular background (yellow) for 40 individual single-cell and single-LD observations, for cells sampled at C/N:15 and C/N:40 at 6, 15, and 100 h (Methods). Specifically, n_{C/N: 15, 6 h} = 8, n_{C/N: 15, 15 h} = 2, n_{C/N: 15, 100 h} = 2, n_{C/N: 40, 6 h} = 12, n_{C/N: 40, 15 h} = 4, n_{C/N: 40, 100 h} = 12. Box-plots represent the 10th, 25th, 75th, and 90th percentile, whiskers represent the 5th and 95th percentile, while the median and mean values of the ensemble distribution are indicated by the horizontal line and sphere, respectively

Fig. 4

Image processing by spatial cross-correlation. a The quantitative-phase image shown in Fig. 2a overlaid with the thresholded areas that exhibit phase-delay values (ΔΦ) comparable to the LDs; the thresholded areas (red) include both parts of the cytosol and the LDs, given their similarity in ΔΦ. b The spatial cross-correlation of the π/2 and π phase-modulated intensity images eliminated the cytosolic background contribution, enabling the error-free localization of the LDs by intensity thresholding (c)

Fig. 5

Comparison between quantitative-mass with conventional imaging. a Scatter plot and marginal histograms of the TAG number density per lipid droplet (LD) as a function of the LD volume for various Y. lipolytica strains and growth conditions (Methods); red line indicates the interquartile range (IQR) for an LD volume of 0.1 µm³, and inset illustrates the number of observations (n), the Spearman correlation coefficient (ρ), yielding p < 0 .001. b A similar scatter plot for the non-TAG cell dry-density as a function of the cell volume with the red line indicating the IQR for a cell volume of 100 µm³; inset illustrates the number of observations (n), the Spearman correlation coefficient (ρ), yielding p < 0.001. The data represent the ensemble of 5 different experimental conditions, each performed in triplicates (see Methods), including: MTYL038_17hr (n_A = 400, n_B = 416, n_G = 535 single-cell observations), MTYL038_28hr (n_A = 664, n_B = 566, n_G = 686), MTYL038_52hr (n_A = 524, n_B = 584, n_G = 762), MTYL038_76hr (n_A = 471, n_B = 825, n_G = 695), MTYL038_100hr (n_A = 633, n_B = 616, n_G = 941), MTYL038_124hr (n_A = 655, n_B = 793, n_G = 894), and Po1g_100hr (n_A = 685, n_B = 748, n_G = 677), yielding a total of 13,770 single-cell and 25,960 single LD observations

Fig. 6

Single-cell growth-productivity trade-offs. a 2D probability distributions illustrating the resource allocation strategies between non-TAG biomass and TAG product for MTYL038 grown at C/N:150 for 17, 100, and 124 h, and Po1g at C/N:150 for 100 h. Each distribution represents the ensemble of three biological replicates, and is portrayed with the color scales noted in the figure. b The cellular noise in non-TAG biomass (red) and TAG product (blue) quantified via the cell-to-cell phenotypic heterogeneity and the robust coefficient of variation (rCV) for MTYL038_17hr (M17), MTYL038_28hr (M28), MTYL038_52hr (M52), MTYL038_76hr (M76), MTYL038_100hr (M100), MTYL038_124hr (M124), and Po1g_100hr (P100). Bars and error-bars indicate the mean and standard-error between three biological replicates, respectively. Under all tested conditions, TAG noise exhibited higher values than growth noise (single-sided t-test between each replicate’s rCV, p < 0.025 for all reported conditions). Further, strong evidence supported that production noise is affected by time following the onset of TAG production (28 h) for MTYL038 (one-way ANOVA, F(4,10) = 4.73 and p = 0.02); no such evidence was detected for growth noise (one-way ANOVA, F(4,10) = 0.92 and p = 0.5). TAG and non-TAG rCV for M100 and P100 were also found to be different (paired-sample t-test, p = 0.05 for both TAG and non-TAG). Inset plots the dependence of the product rCV on the non-TAG biomass rCV, yielding a Spearman correlation coefficient of 0.14 (p = 0.76). Source data of Fig. 6a are provided as a Source Data file

Fig. 7

Strain classification by quantitative mass imaging. a Single-cell productivity distributions and concatenated non-linear fits for MTYL038₁₇ (fit: dF = 56, adj-R² = 0.99, red-x² = 1.71), MTYL038₅₂ (fit: dF = 56, adj-R² = 0.94, red-x² = 3.53), and MTYL038₁₂₄ (fit: dF = 56, adj-R² = 0.95, red-x² = 1.75). Bars and error-bars indicate the mean and standard-error between three biological replicates respectively, while gray arrows indicate the presence of overproducing MTYL038₁₂₄ subpopulations. In red, the 95% confidence band for each concatenated fit is shown. Strong evidence supported the temporal dependence of productivity at these timepoints (one-way ANOVA, F(2,6) = 116 and p < 0.001). b The Kolmogorov-Smirnov (KS) and Euclidian-average distances of the [MTYL038₁₇–MTYL038₅₂] and [MTYL038₁₇–MTYL038₁₂₄] pairs. Bars and error-bars indicate the mean and standard-error between three biological replicates, respectively. Strong evidence supported that the Euclidean-average distances between pairs are different (paired-sample t-test, p = 0.02); no such evidence was observed for the KS distances (paired-sample t-test, p = 0.27). Source data are provided as a Source Data file

Fig. 8

Trade-offs, and heterogeneity under starvation. a The time-dependent change of cell non-TAG biomass, TAG product, and cell-doubling (estimated from bulk optical density measurements, Supplementary Fig. 6) for MTYL038 at C/N: 150. Bars and error-bars indicate the mean and standard-error between three biological replicates, respectively. Using repeated measure ANOVA, a statistically significant effect of time was noted for Δ[TAG] (Greenhouse Geiser adjusted p_GG = 0.05 with adjustment ε_GG = 0.259) for Δ[OD] (p_GG = 0.03 with ε_GG = 0.253), and to a lesser degree for Δ[non-TAG] (p_GG = 0.1 with ε_GG = 0.380). b Decile differences (Δ) computed using the shift function for the TAG product (blue, 100–124 h period) and non-TAG biomass (red, 52–124 h period) as a function of the corresponding decile at 124 h. Asterisks denote statistically significant decile decreases (one-sided t-test, p < 0.04), with bars and error-bars indicating the mean and standard-error between three biological replicates, respectively. Source data are provided as a Source Data file