Origins of Cell-to-Cell Bioprocessing Diversity and Implications of the Extracellular Environment Revealed at the Single-Cell Level

Bioprocess limitations imposed by microbial cell-to-cell phenotypic diversity remain poorly understood. To address this, we investigated the origins of such culture diversity during lipid production and assessed the impact of the fermentation microenvironment. We measured the single-cell lipid production dynamics in a time-invariant microfluidic environment and discovered that production is not monotonic, but rather sporadic with time. To characterize this, we introduce bioprocessing noise and identify its epigenetic origins. We linked such intracellular production fluctuations with cell-to-cell productivity diversity in culture. This unmasked the phenotypic diversity amplification by the culture microenvironment, a critical parameter in strain engineering as well as metabolic disease treatment.

lipids, while the latter was particularly pertinent in single-cell immobilization in microfluidics, addressing the potential of device clogging during exponential growth. Following microfluidic immobilization, cells were occasionally observed to undergo budding during time-lapse confocal imaging with a probability of approximately 10% (Fig. S4); due to the low lipid content and associated low optical signal to noise ratio, such cases were discarded from our study.
Sample Preparation: For both the single-cell analysis in microfluidics and image cytometry, a 1.5 ml cell suspension (1x dilution) was collected following 24 hours of continuous growth. As discussed within the main text, to fluorescently visualize the lipid droplets, a bodipy dye (BODIPY® 493/503 (4,4-Difluoro-1,3,5,7,8-Pentamethyl-4-Bora-3a,4a-Diaza-s-Indacene -Molecular Probes) solution in DMSO (Molecular Probes) was added in YPD media at a concentration of 250 ng/ml, followed by a 4 hour long incubation. For image cytometry, we sampled 50 μL from the cell suspension, deposited them on a coverslip, and covered it on top by a second coverslip to minimize cell motion and out-of-focus contributions. Specific to the microfluidic experiments, following the 4 hour long staining period, the cells were transferred to a gastight syringe at a 10x dilution in YPD media containing a lower concentration of the bodipy dye (100 ng/ml) and 0.02% DMSO and loaded to the microfluidics through one inlet (Fig. S5a).
Following immobilization, the same solution (YPD-Bodipy concentration 100 ng/ml, 0.02% DMSO) was continuously supplied through a second inlet at a rate of 1 μL/min to enable stable immobilization. A 40 ng/ml propidium iodide solution in water (Sigma Aldrich) was included in the media during immobilization to detect cell apoptosis and death. No such phenomena were observed however during the fluctuation analysis experiments. The cells at late stationary phase remained viable for more than 9 hours. This was determined through the propidium iodide dye (Sigma Aldrich) in the media (Fig. S3). 4 Imaging: Vesicle photonics through confocal fluorescent microscopy was employed for visualizing the stained lipid droplets [10]. To this end, an inverted microscope (Leica DMI6000) coupled to a spinning disk confocal system (Yokogawa, CSU10) was employed. A two wavelength laser illumination (488 nm for lipids and 532 nm for viability and apoptosis detection) as well as bright-field imaging (for cell size) were employed. Images were acquired through a 100x oil objective (100x/1.4 NA) on a CCD array (Photometrics CoolSNAP HQ2, pixel size 6.45 µm x 6.45 µm) with typical acquisition settings of a 300 msec integration time, ~10 mW optical excitation on the sample for both lines and a z-scanning step of ~250 nm. Confocal imaging was essential for enhanced imaging contrast and distinguishing lipid droplets in 3D since occasionally LDs overlapped along the optical sectioning path (z-axis) and were thus indistinguishable with conventional epifluorescence microscopy ( Fig. S6). Neutral lipid expression was sampled on average 10 times every 20 min. The lipid content of individual cells (Si) was determined through the ratio of the product between the number and area of cytosolic lipid droplets over the cell size as determined by bright field microscopy.
Each cell required approximately 1-1.5 min of optical sampling for all three wavelength channels.
This imposed an upper limit to the temporal resolution of the lipid expression fluctuation analysis, which we accommodated by selecting a time step of 20 min for the whole cell array. Under such imaging and staining conditions, the smallest detectable lipid droplet was of an area equal to 0.09 µm 2 . This limit was set by requiring a signal to noise ratio > 2 with the background primarily stemming from intracellular fluorescence (Fig. S7). In the context of spatial resolution (~200 nm according to the manufacturer), a single dark pixel criterion was used to distinguish two lipid droplets. 5 Due to the lower axial resolution (z-axis) of the microscope than the lateral one (xy-plane), we quantified the size of the lipid droplets using the maximum intensity projection analysis rather than direct 3D confocal imaging [11]. This technique also enabled sparser optical sampling along the z-axis, thus minimizing bleaching and phototoxicity. A comparison between the maximum intensity projection analysis and direct 3D confocal imaging of lipid droplets yielded that the estimated lipid content per cell scaled linearly between the two methods with an approximate 9% overestimation in the latter case (Fig. S8).
Following acquisition, the images were stored for processing and analysis using ImageJ (National Institutes of Health). For processing, the 3D confocal images were converted to 2D through the maximum intensity projection analysis, followed by convolution with a Gaussian (radius of decay = standard deviation -σ) to attenuate its low spatial-frequency components. Subsequently a bandpass filter was applied to the images (filter size equal to maximum feature size), followed by subtracting the filtered image from the original one (droplet-finder plug-in, ImageJ). The lipid droplets were then identified by applying thresholding based on the histogram entropy criterion [12]. In this way, the number and area of individual lipid droplets were determined and normalized over the cell area as determined by bright field imaging.
Data Analysis: Due to the presence of outliers in our observations (some cells exhibiting an increased lipid content), robust statistics were employed in data analysis (see for example [13]).
To this end, a custom script was written in Matlab for performing the statistical analysis for the longitudinal fluctuations and noise determination. Linear fits and correlation analyses were

Fig. S6
A single Po1g cell imaged with spinning disk confocal microscopy; the cell wall is highlighted through a dotted blue line in the xy plane (maximum intensity projection analysis -left). On the right a section along the optical path (z-axis) is shown, highlight the vertical overlap of one lipid droplet with a manifold of two. 13

Fig. S7
The relationship between the intensity (y-axis) and area (x-axis) of individual lipid droplets (blue dots), including a linear fit with a r = 0.97 Pearson's coefficient (blue line). In the same experiment, the intracellular background fluorescence was determined and shown in red dots. The imaging and staining conditions are discussed in the Supplementary Materials (bodipy dye concentration 100 ng/ml). 14

Fig. S8
The experimentally determined linearity between the single-cell lipid content determined by direct 3D confocal imaging (x-axis) and indirect 2D images through the maximum intensity projection analysis (M.I.P.A.). For the latter, the following experssion was employed to convert 2D areas (circles) to 3D volumes (spheres): volume = 2.4*[area] 3/2 . The linear fit exhibits a Pearson's coefficient of r = 0.99 and a slope of 1.09, indicating a constant 9 % overestimation of lipid content using the maximum intensity projection analysis.