Introduction

Cellular heterogeneity is a fundamental feature of all biological systems. Individual cells within a population exhibit substantial variability in their responses despite being in identical environmental conditions. These responses influence processes such as infections,drug susceptibility, stem cell differentiation and cancer progression (Huang, 2009; Altschuler et al., 2010; Dey et al., 2014; Gough et al., 2014; Gingold et al., 2015; Krieger et al., 2015; Komin et al., 2017). Such variability can arise from stochastic transcriptional bursting (Harper et al., 2011; Rodriguez et al., 2019), variation in chromatin states (Carter et al., 2021; Cheow et al., 2016; Linker et al., 2019) and asymmetry during cell division (Q. Chen et al., 2018). A central question is how such heterogeneity arises and whether it acts as a regulatory mechanism or an emergent consequence of cellular complexity.

Host-pathogen interactions are a striking case of heterogeneity, given the interplay of two dynamic systems. During intracellular infection, clonal host cells can display diverse outcomes ranging from variable infectivity, resistance, containment, clearance, or apoptosis - at the same time (Avraham and Hung, 2016; Delince et al., 2016; S. Helaine et al., 2010; McIntrye et al., 1967; Monack et al., 1996). Whether this diversity reflects intrinsic stochasticity, stable cellular states, or dynamic adaptation remains unresolved, yet it likely shapes population-level infection trajectories. Two nonexclusive models have been proposed for variability in host-pathogen encounters: infection-induced heterogeneity versus pre-existing differences that set trajectories from the outset. Single-cell RNA-seq of Salmonella-infected macrophages supports the latter, showing that pre-existing bacterial expression states influence macrophage responses (Avraham, Haseley, et al., 2015). Variation in macrophage polarization creates intracellular niches that allow Salmonella either to evade detection and persist or to exploit the host for rapid replication (Saliba et al., 2016). Consistent with this view, our prior work showed that pre-existing heterogeneity in host endocytic capacity drives M. tuberculosis (Mtb) infectivity and subcellular trafficking (Sachdeva et al., 2020). Likewise, pre-existing differences among primary human vascular endothelial cells (HUVECs) predict susceptibility to Listeria monocytogenes, with variability in bacterial adhesion as a principal determinant (Rengarajan et al., 2020). Together, these studies illustrate how host and bacterial heterogeneity intersect to generate distinct subpopulations with divergent behaviors. Mtb infection exemplifies this principle, with variability in both pathogen and host contributing to outcomes across cellular and tissue scales (Cadena et al., 2017).

Bacterial phenotypic diversity is advantageous for survival across host environments (S. Helaine et al., 2010; Adams et al., 2011; Sophie Helaine, Cheverton, et al., 2014; Avraham and Hung, 2016; Liu et al., 2016). In Mtb, phenotypic diversification is a well-documented survival strategy, with subpopulations adopting distinct physiological states that favor persistence under stress (Parbhoo et al., 2022; Dhar, J. McKinney, et al., 2016). Reporter strains that sense redox, pH, and ATP fluctuations reveal heterogeneous microenvironments within host cells and tissues (Tan, Sukumar, et al., 2013; Bhaskar et al., 2014; Sukumar et al., 2014; Mehta et al., 2015; Tan, Yates, et al., 2017; Akela et al., 2021), and show substantial variability in bacterial states even within the same host context (Bhaskar et al., 2014). While host cellular and tissue environments clearly influence bacterial states and antibiotic recalcitrance (Bhaskar et al., 2014; Sophie Helaine, Conlon, et al., 2024; Liu et al., 2016), the plasticity of Mtb phenotypic states and the mechanisms at the host–pathogen interface that drive them remain open questions.

Endocytosis is a fundamental cellular process central to nutrient uptake, signaling, and immune defense, which also varies markedly at the single-cell level. Multiple pathways can operate in the same cell (Mayor et al., 2007; Thottacherry et al., 2019), yet individual cells differ in cargo internalization, vesicle trafficking, and intracellular sorting (Dey et al., 2014). A Genome-wide screen has identified genetic modulators that differentially influence endocytosis, with some exerting pathway-specific effects and others broadly regulating multiple uptake mechanisms (Collinet et al., 2010). Rab5, a key early endosomal regulator, coordinates these processes, and its depletion disrupts both endocytosis and lysosomal function (Zeigerer et al., 2012). While some studies suggest that stochastic fluctuations and cell–cell communication shape these differences (Altschuler et al., 2010; Snijder et al., 2009), the origins and drivers of endocytic heterogeneity remain unclear. Given that Mtb primarily enters host cells via phagocytosis and our previous work linking endocytic heterogeneity to Mtb infectivity and subcellular trafficking (Sachdeva et al., 2020), understanding how subpopulation-level differences in endocytosis affect infection dynamics is particularly relevant. Single-cell RNA-seq and live-cell imaging have revealed substantial host and bacterial heterogeneity during infection (Mills et al., 2017; Delince et al., 2016; Mottet et al., 2021), while Mtb stress-reporters (Akela et al., 2021; Bhaskar et al., 2014; Mehta et al., 2015; Sukumar et al., 2014; Tan, Sukumar, et al., 2013; Tan, Yates, et al., 2017) expose coexisting bacterial subpopulations in distinct host niches. Yet, we lack a temporal, causal framework linking specific host cell states to the emergence and dynamics of these bacterial phenotypes.

We integrate live-cell imaging, scRNA-seq, and redox-sensitive Mtb reporter to determine how host cell heterogeneity drives bacterial phenotypes. We show that cell-cycle progression during interphase is a major source of single-cell variation in endocytic capacity, and that these differences drive phenotypic diversity in intracellular Mtb. Cells with distinct endocytic capacity harbor Mtb in different redox states, and experimental manipulation of the cell cycle reprograms both host endocytic profiles and the distribution of bacterial phenotypes. Temporal shifts in endocytic capacity during interphase are mirrored by corresponding changes in intracellular Mtb states, establishing a dynamic, causal link from host cell state to pathogen heterogeneity. Thus, interphase progression regulates endocytic heterogeneity and actively shapes the bacterial phenotypic distribution.

Methods

Cell Culture

RAW 264.7, THP-1 monocytes, and J774 cell lines were cultured in RPMI medium (Gibco, 31800022). HeLa, HEK293FT, immortalised BMDMs, and Vero E6 cells were maintained in DMEM (Gibco, 12800017). Both media were supplemented with 10% fetal bovine serum (heat-inactivated) (FBS; Gibco, 16000044) and 1× Antibiotic-Antimycotic (100×; Gibco, 15240062). THP-1 monocytes were differentiated into macrophages using 20 ng/mL phorbol 12-myristate 13-acetate (PMA; Sigma-Aldrich, P8139) as described in (Baxter et al., 2020). HeLa BAC GFP-Rab5 cells were cultured in complete DMEM supplemented with 500 µM Geneticin (G418; Invitrogen, 10131035). L929 cells were maintained in complete DMEM.cell-conditioned medium was collected from confluent L929 cultures.All cells were maintained at 37^◦C in a humidified atmosphere containing 5% CO₂ and 95% relative humidity.

Isolation of Bone Marrow Derived Macrophages (BMDMs) for infection and functional assays

Monocytes were isolated from the bone marrow of C57BL/6 mice and differentiated into macrophages following a standard protocol (Toda et al., 2021; Nagabhushanam et al., 2003). Briefly, bone marrow was flushed from the tibia and femur using a syringe filled with complete DMEM, and cells were plated in DMEM supplemented with 10% fetal bovine serum (FBS) and 20% L929-conditioned medium. After 7 days of differentiation, cells were washed with PBS and seeded into 12-well plates. Subsequently, these macrophages were used for infection experiments and functional assays, including dextran uptake and cell cycle analysis using DRAQ-5.

Endocytic Uptake Assay

Endocytic uptake assays were performed using fluorescently labeled dextran (Life Technologies, D-1817 and D-22914) or transferrin (Life Technologies, T-2336 and T-23365). Cells were pulsed with dextran (1 mg/mL) or transferrin (5 µg/mL) for 10 minutes in incomplete RPMI or DMEM as described in (Sachdeva et al., 2020). Following the pulse, cells were washed with incomplete RPMI and processed for downstream assays.

Sorting cells into subpopulations based on endocytic capacity

Cells from the parental endocytic population were sorted into low (10–15%) and high (10–15%) endocytic subpopulations based on the fluorescent intensity of the cargo or GFP-Rab5 intensity in a BD FACS Aria Fusion or Aria III sorter. After sorting, cells were analyzed to confirm the accuracy of the fluoresence intensity distribution of the sorted subpopulations. The sorted subpopulations were maintained in RPMI or DMEM media supplemented with 10% fetal bovine serum (FBS) at 37°C in a humidified atmosphere containing 5% CO₂ and 95% relative humidity.

Live cell imaging and image analysis

Live-cell imaging of the ‘low’ and ‘high’ endocytic subpopulations of HeLa BAC GFP-Rab5 cells was performed using a NIKON Ti2 Eclipse epifluoresence microscope equipped with an on-stage incubator (OKO LAB) to maintain physiological conditions (37°C, 5%CO₂, and 95% relative humidity) for 72 hours, with imaging intervals of 30 minutes. We built a customised image analysis pipeline. First, image pre-processing, including background subtraction, was performed using Fiji (Schindelin et al., 2012). Cells were segmented using the GFP fluorescence using CellPose (version 2.3.2) (Stringer et al., 2021; Pachitariu et al., 2022), with a custom segmentation module trained on the Cyto2 model using over 2000 manually curated masks. All segmentation masks were manually verified, and erroneous masks corrected as needed. The resulting cell masks were imported into Ilastik (version 1.4.0) (Sommer et al., 2011) for lineage tracking. All lineage tracks were manually verified. An example of segmentation masks and lineage tracking is provided in Figure S3. Intensity profiles and lineage data were exported from ilastik as CSV files and analyzed in R, using the ggplot2 and dplyr libraries for data visualization and analysis.

Cell synchronization

HeLa BAC GFP-Rab5 cells were seeded in 12-well plates at 60% confluency and treated the following day with 2.5mM thymidine for 18 hours. After treatment, cells were washed to remove thymidine (release) and cultured in complete DMEM for 8 hours. At 0 or 8 hours post-release, cells were pulsed with transferrin-AF647 for 10 minutes, washed three times with incomplete media, trypsinised, and fixed in 80% ethanol. Cell cycle analysis was performed using FlowJo v10.8 following a modified protocol from (Kim et al., 2015). Briefly, ethanol fixed cells were then centrifuged at 3500 RPM (1150*g) for 5 minutes, and resuspended in PBS containing 0.1 % Triton X-100 and Hoechst (5 µg/mL). Cells were centrifuged after 10 minute incubation and resuspended in FACS buffer (1% FBS in 1X PBS) and analysed using BD FACS Fortessa X20 SORP. Cell cycle analysis of RAW 264.7 cells was similar to the above, with the following modifications: thymidine treatment lasted 12 hours, cells were fixed with 4% PFA for 4 hours after release from the block, and then removed from the culture dish by scraping.

Immunostaining, imaging and analysis

RAW 264.7 macrophages and HeLa cells were fixed with 4% paraformaldehyde and permeabilized in SAP buffer (0.2% saponin (Sigma-Aldrich, S4521), 0.2%gelatin (HML, TC041) in PBS) for 10 minutes at room temperature. Cells were then incubated with primary antibodies against EEA1 (1:100; CST #3288 or BD 610457) and RAB5 (1:100; BD 610282) for 2 hours at room temperature. After washing with SG-PBS (0.02% saponin, 0.2%gelatin in PBS), cells were incubated with appropriate secondary antibodies prepared in SG-PBS (1:500) for 1 hour at room temperature, following which cells were washed with SG-PBS and stained with DAPI (1 µg/mL; Invitrogen, D1306) for nuclear visualization. Imaging was performed using either a Nikon Ti2 or Opera Phenix microscope. Image analysis was performed using CellProfiler (Kamentsky et al., 2011) or Harmony software, using pipelines adapted from previous publication (Sachdeva et al., 2020). Briefly, cells were segmented to identify nuclei, cell boundaries, and endosomal compartments. Segmented objects such as nuclei and endosomes were assigned to individual cells, and multiple features—including number, size, and fluorescence intensity—were extracted at single-cell level. Data analysis and visualization were performed in RStudio (Version 2024.04.1+748) using the ggplot2 and dplyr packages.

M.tuberculosis Mrx1-roGFP2 H37Rv Infection

Mycobacterium tuberculosis cultures were grown in Middlebrook 7H9 broth supplemented with Middlebrook OADC Enrichment (BD Biosciences, 211886), 5% glycerol, and 0.05% Tween-80 (Sigma-Aldrich, P4780). For all experiments, bacterial cultures were used in mid-log phase (OD₆₀₀ = 0.6–0.8). Hygromycin B (Invitrogen, 10687010) was added at a final concentration of 50 µg/mL for the maintenance of the M. tuberculosis Mrx1-roGFP2 H37Rv strain.

The M. tuberculosis Mrx1-roGFP2 H37Rv reporter strain was used to assess intracellular redox potential in ‘low’ and ‘high’ endocytic subpopulations. Macrophages (RAW, THP-1 monocyte derived, or bone marrow derived) cells were seeded in 12-well plates at a density of 4 × 10⁵ cells per well and infected with a multiplicity of infection (MOI) of 10 for 2 hours. Cells were washed to remove extracellular Mtb and pulsed with either Transferrin-AF647 or Dextran-TMR for 10 minutes, followed by fixation with N-ethyl maleimide (NEM) for 10 minutes and 4% paraformaldehyde (PFA) for 25 minutes.

Each experiment included calibration controls to measure the response of the Mtb reporter strain to standard oxidative (cumene hydroperoxide, CHP; 5 mM) and reductive (dithiothreitol, DTT; 40 mM) treatments. Samples were analyzed by flow cytometry on a BD LSR Fortessa. Emission from MtbMrx1-roGFP2 was collected at 510 nm following two different excitations at 405 nm and 488 nm, respectively. For each experiment, data were acquired from 10,000 infected cells and analyzed using FlowJo v10.8. For redox measurements of intracellular Mtb in endocytic subpopulations, cells were stratified into the lower and upper 30% of the transferrin or dextran fluorescence intensity distribution, representing ‘low’ and ‘high’ endocytic capacity groups, respectively.

Intracellular redox potential was calculated as previously described in (Bhaskar et al., 2014). Briefly, the emission ratio at the two excitation wavelengths provides a quantitative measure of the redox stress experienced by intracellular Mtb. The degree of oxidation was determined by normalizing the observed fluorescence ratio to values obtained from oxidized (CHP) and reduced (DTT) controls using the following equation:

where R is the observed ratio, R_red and R_ox are the ratios of completely reduced (DTT treatment) and oxidized (CHP treatment) roGFP2, respectively. I 490_minand I 490_max are the fluorescence intensities measured with excitation at 490 nm for fully oxidized and fully reduced roGFP2, respectively. The intracellular redox potential E_roGFP2 was calculated using the Nernst equation:

Based on the midpoint redox potential range of roGFP2 (–270 to –280 mV), gates were defined to determine the proportion of basal, oxidized, and reduced E_MSH Mtbsubpopulations. For each experimental condition, the percentage of cells within each E_MSH subpopulation was calculated and visualized using Microsoft Excel.

To assess bacterial redox states across different cell cycle stages, RAW 264.7 cells were pre-treated with thymidine (2 mM) for 12 hours. Infections were carried out at 0 and 4 hours post-release using Mtb Mrx1-roGFP2 H37Rv at an MOI of 20 for 30 minutes. The high MOI in these experiments was necessitated by shorter infection times due to active cycling of RAW cells. Following infection, cells were fixed with NEM for 10 minutes and then with 4% PFA for 25 minutes. For cell cycle analysis, the standard Hoechst could not be used due to Mtb fluorescence at 405 nm, instead DRAQ-5 (Cell Signaling Technology, #4084) was used at a 1:500 dilution for 10 minutes at room temperature prior to FACS analysis. To assess bacterial redox states between cells at different cell cycle stages, measurements were compared between untreated and thymidine-treated cells.

Single cell RNA sequencing

To check the differential gene expression at single cell level between endocytic subpopulations, RAW 264.7 cells were pulsed with dextran-TMR (1 mg/ml) for 10 mins and sorted into low 10-15%and high 10-15% endocytic subpopulations. In parallel, cells from the same passage were treated with 2 mM thymidine for 12 hrs, the block released and the cells were grown in complete RPMI for 3hrs 30 min. For the single cell sequencing, the sorted low and high dextran subpopulations, along with parental and thymidine treated cells (after 3hrs 30 min of release) were taken as different samples. All the samples were tested for cell viability prior to single cell sequencing. Sequencing was performed using Chromium Next GEM Single Cell 3’LT Kit v3.1 (PN-1000325)(User Guide CG000399). Briefly, cells were resuspended in the RT master mix and loaded into a Next GEM Chip G, along with gel beads and partitioning oil, to generate Gel Bead-In-Emulsions (GEMs) using the Chromium Controller. Within each GEM, cells were lysed and polyadenylated RNA was reverse transcribed, incorporating a template switch oligo and barcoded oligo-dT primers that included an Illumina Read 1 (R1) primer sequence, a 10x cell barcode, and a Unique Molecular Identifier (UMI). Following reverse transcription, GEMs were broken and the pooled, barcoded cDNA was purified using Silane DynaBeads, then amplified by PCR. Size selection of appropriately sized cDNA fragments was performed using SPRIselect beads. During library construction, enzymatic fragmentation, end repair, and A-tailing were followed by adapter ligation and PCR amplification to add the Illumina Read 2 (R2) primer site, P5/P7 flow cell adapters, and a sample index. The final libraries were assessed for quality and size prior to sequencing on an Illumina platform. The data were analysed following current best practices for scRNA-seq analysis ( (Heumos et al., 2023) as detailed below.

scRNA seq Data Analysis

Single-cell RNA-seq Preprocessing

Raw FASTQ files were processed using Cell Ranger v7.2.0 (10x Genomics) on a Linux system. The software package (cellranger-7.2.0.tar.gz) and the mouse reference genome (refdata-gex-mm10-2020-A.tar.gz) were extracted using the tar command. The reference transcriptome used was Mus musculus mm10 (2020-A release). The cellranger count pipeline (v7.2.0, 10x Genomics) was run individually for each sample using appropriate parameters to process demultiplexed FASTQ files against the mm10-2020-A reference genome. This step performed alignment, barcode and UMI processing, and gene expression quantification, generating gene–barcode matrices and associated quality control metrics for downstream analysis.

Data Import and Construction of AnnData objects

To enable downstream single-cell RNA-seq analysis in Python, the Cell Ranger output matrices (matrix.mtx.gz, features.tsv.gz, and barcodes.tsv.gz) were parsed and converted into a gene-by-cell expression matrix. This matrix was saved as a CSV file and subsequently used to construct an AnnData (.h5ad) object, incorporating both gene and cell metadata, facilitating further analysis using the Scanpy framework (Wolf et al., 2018).

Quality Control in Python

Quality control (QC) metrics were calculated to assess cell quality across all conditions by identifying mitochondrial genes (prefix mt-) in gene annotations. Using Scanpy’s calculate qc metrics function, we computed standard QC metrics, including the proportion of mitochondrial reads, the number of highly expressed genes, and logarithmically transformed features. This ensured the proper identification of cells with potential mitochondrial contamination.

In addition to cell-level filtering, genes were filtered based on their expression levels. Specifically, only genes expressed in at least 20 cells were retained for downstream analysis. This step reduces technical noise and focuses the analysis on biologically relevant signals.

Violin plots were generated for key QC metrics: number of genes detected (n genes by counts), total counts (total counts), and percentage of mitochondrial genes (pct counts mt) for both high and low conditions. These visualizations helped identify low-quality cells exhibiting characteristics such as low gene detection, high mitochondrial content, or outlier total counts.

Based on these plots, cells with excessive mitochondrial contamination (greater than 10%) and outliers in gene count or total counts were filtered out to improve overall data quality.

Normalization and Log Transformation

To enable comparison between conditions, a new column labeled Condition was added to the obs metadata in all datasets. The data were then normalized using the normalize total function, scaling counts per cell to 10⁴ to account for varying sequencing depths. A log transformation (log1p) was applied to stabilize variance and reduce the influence of highly expressed genes. Finally, the processed datasets were saved as .h5ad files for future analysis.

Data concatenation and Pseudobulk generation

After quality control, the datasets to be analysed for differential gene expression were merged into a single AnnData object using anndata.concat() along the cell axis, with a new Condition column added to indicate the condition (“high” or “low”). The combined dataset was saved as combined quality control.h5ad for downstream analysis. For pseudobulk analysis, the count matrix was extracted from the combined AnnData object as a pandas DataFrame with cell barcodes made unique and condition labels retained. To generate the pseudobulk data, cells from each condition were randomly grouped into three pseudoreplicates using a fixed random seed for reproducibility. The resulting annotated matrix, containing both Condition and Replicate columns, was saved along with the corresponding cell-level metadata and gene-level metadata for use in downstream analyses.

Data Aggregation and Dimensionality Reduction

The data was then aggregated by summing expression values across all cells within each replicate for both conditions, creating replicate-level profiles. These profiles were saved as new AnnData objects and later combined into one dataset.

Differential Gene Expression Analysis

To identify differentially expressed genes (DEGs) between conditions, the pseudobulked AnnData object was analyzed using the edgeR package in R via the rpy2 interface. Genes with a false discovery rate (FDR) less than 0.05 were considered significant. All intermediate R objects and results, including the fitted model, DEG list, design matrix, and statistical results, were saved as RDS files to ensure reproducibility and facilitate future analysis.

Single cell colony sorting and expansion

Single-cell sorting was performed using a Beckman CytoFLEX SRT three-laser system. HeLa GFP-RAB5 cells were cultured to confluency, and GFP-positive cells were gated using wild-type HeLa cells as a negative control. Individual GFP-positive HeLa cells were sorted into 384-well optical bottom plates containing 80 µL of DMEM supplemented with 10% FBS and 1× Antibiotic-Antimycotic (100×). Following sorting, cells were cultured for 3 days, after which the media was replenished. After one week, wells with visible cell colonies were trypsinized and expanded into 48-well plates and subsequently into 25 cm² flasks. Once sufficient expansion was achieved, stocks for each clonal population were generated from cells cultured in T25 flasks. These clonal stocks were then used for downstream experiments.

Determination of Asymmetry

To determine the coefficient of asymmetric binomial partitioning of GFP-Rab5 in the daughter cells from the time-lapse data, cells were classified into mothers and daughters. The analysis was based on the relationship of the variance in intensity of daughter cells predicted from the statistics of the binomial distribution process (Peruzzi et al., 2021).

where σ_d is the standard deviation of GFP RAB5 intensities in the daughter cells just after birth, a is a rescaling variable to rescale the GFP RAB5 intensities by 10⁻⁶, p is the probability of partitioning, µ_m and σ_m is mean and standard deviation of the scaled GFP RAB5 intensity in the mother cells just before division. Cells tracked for 5 hours or more were selected for analysis. For these cells, mother and daughter identities were assigned. The mean and standard deviation of total GFP-Rab5 intensity (TI) were calculated separately for mothers and daughters. For mothers, the TI value was taken 30 minutes prior to cell division; for daughters, the TI value was recorded at the birth timepoint (i.e., immediately after division). All calculations were carried out independently for each of the three strains (High, Parent, Low).

To estimate the asymmetric binomial partitioning coefficient p, a bootstrapping approach was employed. TI data were sampled with replacement 1000 times. In each iteration, a mean and standard deviation of TI were computed for mother and daughter cells, which were then used to calculate a value of p using the partitioning equation (Eq.3). Repeating this over 1000 iterations produced a distribution of p values.

Comparison of Inter-mitotic Times (IMT) of cells across strains and generations

Cells in different strains were segregated into generations based on frequency of the divisions. Starting cells were categorized as generation ‘zero’, their progeny were subsequently termed as ‘first’, ‘second’ and so on. For the generation zero cells, the time to their division was extracted and used as a proxy for IMT. For the rest of the generations, the time interval between birth and the division event was used to calculate the IMT. Then, the IMT of cells belonging to different strains was compared for all the generations. A non parametric Wilcoxon Rank Sum Test was performed to compare the distributions as the data was non-Gaussian with long tails.

Linear regression for calculating slopes of time-series data

Time-lapse imaging of HeLa BAC cells provided repeated measurements of GFP–Rab5 intensity in the same cells over time under different experimental conditions. Since total GFP intensity increased monotonically, we applied a linear model (lm function in R) to estimate the slopes. For untreated and post–thymidine block cells, track lengths were trimmed to match the thymidine block. Although the block itself was 18 h, we used a 20-h analysis window to allow 2 h for cell recovery. Tracks shorter than 3.5 h were excluded. Data analysis and plotting were carried out in KNIME and R.

Variance decomposition analysis

The contribution of a random variable X to the total variance of a random variable Y can be computed as follows:

where the first term on the right hand side denotes the variance explained by X, while the second term corresponds to the remnant variation in Y that cannot be explained by X. In our context, Y is either GFP-Rab5 or Transferrin intensity, while the covariate X is Hoechst intensity. Channel values from flow cytometry data was used for these calculations.

Results

Heterogeneity in endocytic capacity promotes Mtb phenotypic diversity

Our previous work showed that macrophage subpopulations with higher endocytic capacity (‘high’ cells) are more permissive to Mtb infection than those with lower capacity (‘low’ cells). Moreover, Mtb trafficking differs between these groups, with bacteria in ‘high’ cells more frequently delivered to lysosomes (Sachdeva et al., 2020), suggesting that they may encounter greater lysosome-associated stress. Such variation in endocytic capacity could therefore generate phenotypic diversity in Mtb. To test this, we infected RAW macrophages with Mtb expressing Mrx1-roGFP2, a redox-sensitive reporter that enables fluorescence-based measurement of oxidative stress associated with phago-lysosomal compartments (Bhaskar et al., 2014). Based on the ratio of bacterial fluorescence intensities at 405 nm and 488 nm excitation, infected cells were classified as harboring E_MSH-‘oxidized,’ ‘reduced,’ or ‘basal’ Mtb phenotypes (Bhaskar et al., 2014). Following infection, cells were pulsed with fluorescently tagged dextran or transferrin - endocytic cargoes with distinct uptake routes and fates (Sachdeva et al., 2020). Fluorescence intensities of these cargoes were then used to define ‘high’ or ‘low’ endocytic capacity subpopulations, and the infectivity and redox state of intracellular Mtb were assessed in these groups (schematic, Figure 1A). Consistent with our previous findings (Sachdeva et al., 2020), the ‘high’ endocytic capacity subpopulations showed higher infectivity (Figure 1B). Importantly, these subpopulations also harbored a significantly greater proportion of E_MSH-oxidized Mtb compared to ‘low’ endocytic capacity subpopulations. Conversely, the proportion of cells harboring E_MSH-reduced Mtb was significantly lower in the ‘high’ group (Figure 1B; Figure S1A). Similar results were observed in infections of THP-1 monocyte-derived macrophages (Figure 1C) and primary mouse bone marrow–derived macrophages (Figure 1D). These findings indicate that intracellular Mtb in cells with higher endocytic capacity experience greater oxidative stress than in those with lower capacity. Across independent experiments and infection systems, high endocytic capacity was consistently associated with a decrease in the E_MSH-reduced subpopulation and a corresponding increase in the E_MSH-oxidised subpopulation, indicating a robust redistribution from reduced to oxidised states. The E_MSH-basal fraction showed only minor changes, suggesting that high endocytic capacity primarily drives exit from the reduced state and promotes oxidised phenotypes without substantially altering the transitional pool. Together, these results demonstrate that host endocytic capacity shapes phenotypic heterogeneity and stress adaptation in intracellular Mtb.

Figure 1:

Characterization of variance in the endocytic capacity across cell types

Our findings raised the question of why endocytic capacity itself varies across cells. To address this, we next investigated the sources of this variation as a cell-intrinsic property, independent of infection. Rab5, a key regulator of endocytosis (Zeigerer et al., 2012), is essential for early endosome formation. Overexpression of Rab5 accelerates both fluid-phase and receptor-mediated endocytosis, establishing it as a rate-limiting factor in this pathway (Bucci et al., 1992). Based on this, we hypothesized that total Rab5 levels determine the endocytic capacity of a cell. To test this, we sorted RAW macrophages into ‘high’ and ‘low’ subpopulations based on dextran uptake (Figure 2A) and immunostained for Rab5 and its effector, Early Endosomal Antigen 1 (EEA1). Image analysis revealed that the number and intensity of Rab5- and EEA1-positive endosomes co-varied with dextran uptake in the ‘high’ and ‘low’ subpopulations (Figure 2B). Thus, early endosomal content, together with cargo uptake, defines the endocytic capacity of distinct subpopulations. We then asked whether this capacity is stable or dynamic over time. Sorted RAW macrophage subpopulations were cultured for three days, and their endocytic capacity was re-analyzed. Notably, the initially distinct ‘high’ and ‘low’ populations converged over time (day 0 to day 3), becoming indistinguishable from the parental distribution (Figure 2C). Similar results were obtained in Raw cells infected with Mtb (Figure 6D i, iii). These findings demonstrate that endocytic capacity is not fixed but a dynamic property of individual cells.

Figure 2:

Figure 3:

We next sought to characterize the variation in endocytic capacity across other cell types. A panel of cell lines was pulsed with fluorescently tagged transferrin or dextran, and cell-to-cell variability was quantified by calculating the coefficient of variation in cargo uptake across individual cells within each population (Figure 2D,E). All tested cell lines displayed a high degree of variability, indicating substantial differences in the endocytic capacities of individual cells. Notably, the coefficient of variation for dextran, a fluid-phase cargo, was consistently higher than that for transferrin, a receptor-mediated cargo, underscoring the more regulated nature of receptor-mediated endocytosis (Mayor et al., 2007). Furthermore, the two uptake modes exhibited variable correlations across different cell types (Figure S2A), suggesting cell type–specific mechanisms governing cargo internalization. Despite these differences, the consistently high variability observed for both cargoes across diverse cell lines demonstrates that heterogeneity in endocytic capacity is a fundamental property of cell populations.

Similar to macrophages, HeLa cells displayed high cell-to-cell variability in endocytosis. To test whether early endosomal content and cargo uptake were similarly associated, as observed in RAW cells, we pulsed HeLa cells stably expressing GFP–Rab5C under its endogenous promoter and regulatory elements (Poser et al., 2008) with fluorescently tagged transferrin or dextran and measured the correlation between cargo uptake and GFP–Rab5 levels. A strong positive correlation was observed for both cargoes (Figure S2B). To further characterize endocytic capacity, HeLa GFP–Rab5 cells were sorted into ‘high’ and ‘low’ subpopulations based on the GFP–Rab5 expression. Immunostaining confirmed corresponding differences in the number of endosomes and the total cellular intensities of Rab5 and its effector EEA1 between the two groups (Figure 2F). We next asked whether these ‘high’ and ‘low’ subpopulations were stable or dynamic, as previously seen in RAW macrophages. After three days in culture, the initially distinct subpopulations had converged, becoming indistinguishable from the parental distribution (Figure 2G). To further confirm this dynamic behavior, single-cell clones were derived from the pool of HeLa BAC GFP–Rab5 cells. In five independent clones, the initially distinct ‘high’ and ‘low’ endocytic capacities again converged to the parental distribution within four days (Figure S2C). Collectively, these findings demonstrate that endocytic capacity is a dynamic property of individual cells and that such variation is broadly conserved across mammalian cell types.

Single cell analysis for drivers of heterogeneity of endocytic capacity

To better resolve the dynamic nature of endocytic capacity at the single-cell level, we performed live-cell time-lapse imaging of HeLa GFP–Rab5 cells (Figure 3A, Movie S1A) for 72 hours at 30-min intervals. A custom image analysis pipeline was developed to segment and track single cells across multiple generations, combining Fiji (Schindelin et al., 2012), CellPose (Stringer et al., 2021; Pachitariu et al., 2022), and Ilastik (Sommer et al., 2011) (Figure S3A, Movie S1B). Since total GFP-Rab5 intensity correlates with cargo uptake (Figure 2E), we used it as a metric for assessing the endocytic capacity of individual cells. Intensity profiles of single cells showed a steady increase in GFP-Rab5 levels over the cell’s lifespan, followed by a reset upon cell division (Figure 3C). To test this behaviour across the endocytic subpopulations, we sorted cells into ‘low’ and ‘high’ subpopulations based on the GFP-Rab5 expression levels and performed live-cell imaging.

We then examined the distribution of total GFP-Rab5 intensity and mean intensity (area normalized) at the start of the imaging period and at defined time intervals (Figure 3B). Analysis of the distribution of these parameters over a large number of cells showed that these parameters were initially distinct between ‘high’ and ‘low’ subpopulations, but eventually converged over time as the cells progressed through the cell cycle. Similar results were obtained in an unsorted population, where ‘high’ and ‘low’ groups were defined from the first frame of imaging (Figure S3B), ruling out cell sorting–associated artifacts. These findings were further corroborated by tracking GFP–Rab5 intensity across cell lineages over multiple generations (Figure 3D). Together, live-cell imaging results are consistent with flow cytometry snapshots (Figure 2G), reinforcing that endocytic capacity is dynamic: initially distinct subpopulations converge toward the parental distribution over successive cellular generations.

We next investigated potential causes of the heterogeneity in endocytic capacity observed in these images. One possibility is asymmetric distribution of organelles during mitosis, which has been reported to contribute to cell-to-cell variability (J. T. Chang et al., 2007; A. Y. Chang et al., 2017; Loeffler, Wehling, et al., 2019; Loeffler, Schneiter, et al., 2022). To test whether inheritance of GFP–Rab5 differed between the ‘high’ and ‘low’ cells, we measured GFP–Rab5 intensity in daughter cells immediately after division and assessed the probability of symmetric inheritance in the two subpopulations. The results (Figure 3D) showed that GFP–Rab5 inheritance was largely symmetric in both ‘high’ and ‘low’ cells, ruling out asymmetric partitioning of GFP–Rab5 as a major contributor to population-level heterogeneity.

We next examined whether differences in inter-mitotic times could account for variation in endocytic capacity. To this end, we analyzed the inter-mitotic times of the ‘high’ and ‘low’ endocytic subpopulations across successive generations (Figure 3F). The first—but not the subsequent—inter-mitotic time was markedly shorter for ‘high’ cells compared to ‘low’ cells. In two independent experiments, division time decreased from 16.16 h to 8.61 h (4̃7% reduction) and from 18 h to 8.5 h (5̃3% reduction), respectively. These findings indicate that immediately after sorting, ‘high’ endocytic cells were enriched in a population closer to mitosis than ‘low’ cells. Thus, sorting based on GFP–Rab5 intensity appears to preferentially enrich for cells in the G2 phase of the cell cycle, consistent with elevated GFP–Rab5 levels observed in single-cell traces preceding mitosis. Taken together, these results suggest that cell-cycle progression underlies heterogeneity in endocytic capacity, generating variability between individual cells at a given time snapshot.

Cell cycle progression is a major driver of heterogeneity of endocytic capacity

Although live-cell imaging suggested a progressive increase in GFP–Rab5 intensity during the cell cycle, the stability of GFP raises the possibility that this trend reflects cumulative accumulation rather than active regulation. To address this, we pulsed HeLa GFP–Rab5 cells with fluorescently labeled transferrin and measured both GFP–Rab5 expression and transferrin uptake in relation to DNA content by flow cytometry. Both parameters showed a significant positive correlation with DNA content (Figure 4A). Imaging assays similarly revealed correlations between DAPI staining and early endosomal content, assessed by GFP–Rab5 and EEA1 (Figure 4B); the lower correlation observed in imaging may reflect measurement in a single plane rather than the full cell volume. Classification of cells into G1, S, and G2 phases confirmed that G2-phase cells display significantly higher GFP–Rab5 levels and transferrin uptake compared to G1 and S-phase cells (Figure 4C). Together, these results indicate that the increase reflects a bona fide, cell cycle–dependent rise in endocytic capacity rather than an artifact of GFP stability.

Figure 4:

To confirm these findings, HeLa GFP-Rab5 BAC cells were arrested at the G1/S boundary using thymidine treatment. Synchronous progression into G2 following release from the block enabled assessment of endocytic capacity in a G2-enriched cell population (Figure S4A). This experiment confirmed that an increased proportion of G2-phase cells leads to higher overall GFP–Rab5 intensity and transferrin uptake in the population (Figure 4D); similar results were obtained in two independent single-cell clones (Figure S4B,C). Further, thymidine-treated cells were immunostained for EEA1 and imaged, revealing a higher abundance of Rab5 and EEA1-positive endosomes in the G2-enriched condition following thymidine release (Figure 4E). Taken together, these results demonstrate that endocytic capacity fluctuates dynamically during interphase in HeLa cells, increasing as cells progress from G1/S to G2 phase.

We next investigated whether the relationship between cell cycle progression and endocytic capacity is conserved in macrophages by pulsing RAW macrophages with fluorescently labeled transferrin and assessing transferrin uptake in subpopulations stratified by DNA content. The results (Figure 4F) show that, similar to HeLa cells, RAW macrophages with higher DNA content exhibit increased endocytic capacity. To further validate this, cells were arrested with thymidine, briefly released from the block, and pulsed with fluorescently labeled dextran. This treatment enriched for G2-phase cells (Figure S4D), which displayed significantly higher dextran uptake (Figure 4G). Collectively, these findings demonstrate that as interphase cells progress from G1 to G2, endocytic uptake and early endosomal abundance increase continuously and reset after mitosis. Thus, cell cycle progression generates single-cell heterogeneity in endocytic capacity during interphase.

Active control of endocytic capacity by interphase progression

To investigate sources of variation in endocytic capacity in macrophages using an independent approach, we performed single-cell RNA sequencing (scRNA-seq) of sorted ‘high’ and ‘low’ endocytic capacity RAW macrophage subpopulations (Figure 5A). The data were analyzed following current best practices for scRNA-seq analysis (Heumos et al., 2023) (Figure 5B). Gene ontology analysis of the top differentially expressed genes between the ‘high’ and ‘low’ subpopulations (Figure S5A) revealed significant enrichment of genes associated with cell division, mitosis, and G2-phase processes in the ‘high’ capacity cells. In contrast, the ‘low’ endocytic capacity subpopulation showed upregulation of genes related to transcriptional regulation, a process typically elevated in G1 phase (Bertoli et al., 2013) (Figure 5C,D). Comparison with cells following thymidine treatment, which enriches for G2-phase cells (Figure S5B), showed greater overlap of genes with the ‘high’ endocytic cells than with ‘low’ cells (Figure S5C). These findings provide additional evidence linking cell cycle progression to endocytic capacity. Further analysis of the scRNA-seq data revealed potential mechanisms beyond cell cycle contributing to endocytic heterogeneity. We overlaid differentially expressed genes (DEGs) between ‘high’ and ‘low’ endocytic subpopulations with DEGs from thymidine-treated versus parental populations (Figure 5Ei). As expected, cell cycle emerged as the most significantly enriched pathway among the intersecting genes (Figure 5Eii). We reasoned that the set of 205 non-overlapping genes could represent cell cycle–independent contributors to endocytic capacity. KEGG pathway analysis of this gene set showed significant enrichment in metabolic and lysosomal pathways. Together, these results establish cell cycle progression as a key determinant of endocytic capacity and identify additional cellular pathways contributing to its heterogeneity.

Figure 5:

While these results demonstrate a relationship between interphase progression and endocytic capacity, they do not exclude the possibility that endocytic capacity might increase constitutively in interphase cells, independent of active cell cycle–mediated regulation. To distinguish between these possibilities, we imaged HeLa GFP–Rab5 BAC cells under thymidine block. As an independent method to manipulate the cell cycle, cells were also grown in low FBS, a condition known to slow cell cycle progression (Orren et al., 1995; Qiuyan Chen et al., 2004). If endocytic capacity increases constitutively (i.e., independently of cell cycle progression), GFP–Rab5 intensity should continue to rise at a consistent rate despite cell cycle arrest. Conversely, if endocytic capacity is directly regulated by cell cycle progression, a decrease in the rate of GFP–Rab5 accumulation would be expected. To evaluate this, we fitted the GFP–Rab5 intensity from time-lapse measurements and calculated the slopes of these fits (Figure 5Fi). Comparison of slope distributions between non-synchronized and synchronized conditions (Figure 5Fii) revealed a significant reduction (∼30%) in the slope during both thymidine block and low FBS treatment. This indicates that cell cycle progression plays a direct role in modulating the heterogeneity in endocytic capacity. To further confirm this, we imaged the same cell population during and after release from thymidine block and assessed the slopes of the total GFP intensity. The results (Figure 5Fiii) showed that the initially low slope increased by ∼30% as cell cycle block was lifted and cells resumed growth. These findings confirm active control of total GFP–Rab5 content and endocytic capacity by cell cycle progression.

To quantify the precise contribution of cell cycle progression to the variation in endocytic capacity, we performed variance decomposition analysis on flow cytometry data (Figure 4D), as detailed in the Methods. The results (Figure 5G) show that approximately 30% of the variance in endocytic capacity - measured by either GFP-Rab5 content or transferrin uptake - can be explained by Hoechst staining, which serves as a marker of cell cycle progression. Thus, nearly one-third of the observed heterogeneity in endocytic capacity is attributable to interphase progression.

Cell cycle progression promotes phenotypic diversity of Mtb

The identification of cell cycle progression as a key determinant of endocytic capacity suggests that it may likewise shape the phenotypic states of intracellular Mtb. To test this, we examined whether the phenotypic states of Mtb Mrx1-roGFP2 vary with the cell cycle phases of the infected host cell (Figure 6A). Asynchronous RAW cell populations were categorized into G1/S or G2 phases based on their DNA content, with gating performed using Draq-5 staining (Figure S6). Consistent with our earlier observations that G2-phase cells display higher endocytic capacity, and with prior report that cells with higher endocytic activity are more permissive to Mtb infection (Sachdeva et al., 2020), G2-phase cells exhibited higher infection rates than G1/S-phase cells (Figure 6Bi). Notably, a greater proportion of Mtb in G2-phase cells was oxidized, while the fraction in the reduced state decreased (Figure 6Bi). Together, these findings indicate that host cell cycle progression, by modulating endocytic capacity, influences the phenotypic state of intracellular Mtb.

Figure 6:

To further validate this, cells were synchronized using thymidine block followed by release to enrich for G2-phase cells, infected with Mtb Mrx1-roGFP2, and the proportions of cells harboring oxidized, reduced, or basal Mtb were quantified. The results (Figure 6Bii) showed an increased proportion of cells with oxidized Mtb following thymidine treatment and in the G2-enriched subpopulation. Together, these data establish that cell cycle–dependent changes in endocytic capacity promote phenotypic diversity of intracellular Mtb.

Our previous result (Figure 3C) showed that endocytic capacity is reset as cells progress through mitosis (M phase). We therefore asked whether Mtb in ‘high’ endocytic capacity cells—experiencing higher oxidative stress—exhibits phenotypic plasticity as the host cell’s endocytic capacity changes following cell division. To test whether Mtb phenotypes adapt as host endocytic capacity changes, infected cells were sorted into ‘high’ and ‘low’ endocytic subpopulations, cultured separately for three days, and bacterial redox states were measured (Figure 6C). Once the initially distinct endocytic distributions had merged (Figure 6Di, iii), the proportion of oxidized Mtb in the former ‘high’ subpopulation decreased to parental levels, and no significant differences remained between the ‘high’ and ‘low’ groups (Figure 6Dii, iv). These results demonstrate that Mtb phenotypes exhibit plasticity in response to dynamic changes in host endocytic capacity as they transition from the G2 phase of the mother cell to the G1 phase of the daughter cells.

Cell cycle driven changes in endocytic capacity and Mtb phenotypic diversity in post-differentiated macrophages

We next investigated whether the host cell cycle and endocytic capacity similarly shape Mtb phenotypes in terminally differentiated macrophages. First, we examined whether the relationship between endocytic capacity and cell cycle stage observed in RAW cells is conserved in THP-1 monocytes and the macrophages derived from them (schematic, Figure 7A). THP-1 monocytes were pulsed with fluorescently labeled dextran and stained with Hoechst to distinguish cells in G1/S and G2 phases of the cell cycle. Endocytic capacity was quantified within these defined subpopulations. The same monocyte population was differentiated into macrophages using PMA stimulation, followed by analysis of both endocytic capacity and cell cycle distribution. The results (Figure 7B) show that the proportions of G1/S and G2 cells remain comparable before and after differentiation, indicating that the cell cycle distribution of the monocyte population is largely maintained in the derived macrophages, consistent with previous reports (Gazova et al., 2020). The slight increase in G2-phase cells within macrophages may result from S-phase cells entering G2 during differentiation. Importantly, the relative differences in endocytic capacity between G1/S and G2 subpopulations persist following differentiation (Figure 7B), demonstrating that cell cycle stages continue to influence endocytic capacity in differentiated cells.

Next, we assessed whether the phenotypic states of Mtb in differentiated macrophages exhibit similar dependence on endocytic capacity and cell cycle stage as observed in cycling cells. Macrophages (THP-1 or BMDM) infected with Mtb Mrx1-roGFP2 redox reporter were pulsed with fluorescently tagged dextran to delineate host subpopulations with ‘high’ or ‘low’ endocytic capacities, and stained with Draq5 to determine DNA content and cell cycle stage (Figure S7). The distribution of Mtb redox phenotypes was then assessed across these host subpopulations. The results (Figure 7C–F) reveal that, similar to cycling cells, macrophages with higher endocytic capacity exhibit greater Mtb infection rates and harbor a higher proportion of oxidized Mtb compared to subpopulations with lower endocytic capacity (Figure 7C, E). Likewise, cells in the G2 phase are more infective and contain a greater proportion of oxidized Mtb than those in G1 (Figure 7D, F). We further stratified the data based on oxidized or reduced Mtb subpopulations and analyzed the distribution of host macrophages across cell cycle stages and endocytic capacities. The results (Figure 7G–J) show that oxidized Mtb predominantly associate with macrophage subpopulations exhibiting higher endocytic capacity, whereas reduced Mtb are more frequently found in cells with lower endocytic activity (Figure 7G, I) in THP-1 monocyte derived macrophages as well as bone marrow derived macrophages. Similarly, reduced Mtb are enriched in G1-phase macrophages, while G2-phase macrophages predominantly harbor oxidized Mtb (Figure 7H, J). These findings demonstrate that despite the absence of active cycling, terminally differentiated macrophages retain cell cycle stage–associated differences in endocytic capacity and influence corresponding Mtb phenotypic states.

Figure 7:

Discussion

Cellular heterogeneity is a hallmark of biological systems (Altschuler et al., 2010; Komin et al., 2017), yet the mechanisms generating this variability and its impact on host-pathogen interactions remain incompletely understood. Here, we define a causal axis linking intrinsic heterogeneity in a fundamental host cellular process to intracellular bacterial phenotypic diversity. Using live-cell imaging, single-cell analyses, and bacterial reporter assays, we show that temporal shifts in endocytosis across interphase stages create structured single-cell variability, generating distinct intracellular niches that drive intracellular Mtb phenotypic adaptation.

These findings build upon our previous observation that cell-to-cell variability in endocytic capacity influences Mtb infectivity and subcellular trafficking (Sachdeva et al., 2020). We now demonstrate that this variability also shapes the intracellular microenvironments encountered by Mtb and alters its phenotypic adaptation (Figure 1). We systematically characterize the distribution and temporal dynamics of endocytic capacity across cellular subpopulations in uninfected cells and show that this is not a static trait but fluctuates over time within individual cells (Figure 2). This temporal variability suggests the presence of a structured source of heterogeneity (Huang, 2009; Movasat et al., 2025). Through live-cell imaging and single-cell RNA sequencing, we identify cell cycle progression as a key intrinsic determinant of endocytic capacity (Figure 3). Cells in early and late interphase, i.e., G1 and G2 stages, exhibit marked differences in their endocytic capacities, and this variance aligns with changes in Mtb uptake and subsequent phenotypic adaptation during these stages. Our data suggest that the host cell cycle acts as a regulatory axis modulating the functional state of the endocytic machinery (Figure S4 and 5), thereby creating temporally dynamic intracellular niches for invading pathogens. This link between interphase progression and endocytic control has striking consequences for the pathogen. We find that Mtb phenotypic diversity is significantly influenced by the host cell’s position in the cell cycle at the time of infection (Figure 6). Notably, this phenomenon persists in post-differentiated macrophages, indicating that the underlying regulatory architecture is preserved beyond cycling precursor populations (Figure 7). Our data suggest that cell-cycle history leaves a lasting imprint and continues to shape cellular behavior in mature innate immune populations even after differentiation. This view is consistent with recent evidence indicating that macrophage functional diversity is maintained through both proliferative history and microenvironmental cues, rather than being erased by terminal differentiation. (Cao et al., 2024; Ahmad et al., 2022; Lavin et al., 2014). Together, these findings position cell cycle progression as a key regulator of functional heterogeneity in macrophages and identify endocytic capacity as a key effector of this variability. This dynamic heterogeneity, in turn, creates a mosaic of intracellular environments that drive bacterial phenotypic diversity.

Our findings resonate with several converging themes in cell and infection biology. They provide mechanistic insight into how functional heterogeneity arises within a cell population. While transcriptional noise, asymmetric division, and epigenetic regulation have been proposed as contributors to cellular heterogeneity (Carter et al., 2021; Cheow et al., 2016; Linker et al., 2019; Q. Chen et al., 2018; Harper et al., 2011; Rodriguez et al., 2019), our live-cell imaging and synchronization experiments identify interphase progression as a strong driver of variation in endocytic capacity. This supports emerging views of the cell cycle as a primary organizer of cellular state and function (Movasat et al., 2025; McDavid et al., 2016; Gruenheit et al., 2018). Our scRNA-seq data raises an intriguing possibility to explain active regulation of endocytosis by cell cycle progression: as cells prepare for volumetric doubling, membrane biogenesis must scale accordingly to account for expansion of the plasma membrane (Raucher et al., 1999) as well as internal compartments such as the ER. We speculate that endocytosis may act as a homeostatic buffer to calibrate plasma membrane composition and tension during volumetric expansion in interphase, thereby linking lipid biosynthesis, membrane dynamics, and trafficking capacity to cell cycle progression. Supporting this model, lipid biosynthesis genes are specifically upregulated in cells with high endocytic capacity.

Bacterial heterogeneity is a known driver of infection outcomes (Sherry et al., 2024; Aldridge et al., 2012; Dewachter et al., 2019), yet the contribution of host cell variability remains underexplored. Here, we establish that host endocytic capacity, which fluctuates with the cell cycle, is a critical factor shaping the intracellular niche. This capacity peaks in G2-phase, rendering cells more infectible. Crucially, we find that Mtb residing within these G2 cells display a more oxidized redox state, indicating they experience significant intracellular stress. Our scRNA-seq analysis provides the molecular explanation for this: high-uptake cells are characterized by an enriched lysosomal gene signature. This host-mediated pressure forces bacterial adaptation, structuring the phenotypic heterogeneity essential for persistence (Goossens et al., 2020; Mishra et al., 2019; Mavi et al., 2020). Our data demonstrates a remarkable phenotypic plasticity in the bacteria; as host cells divide and their daughters enter G1, their endocytic capacity resets, and we observe a corresponding reversion in the Mtb redox state. This shows that the host cell cycle continuously and dynamically structures the bacterial phenotypic diversity.

These results suggest that host cellular processes generate distinct cellular subpopulations that promote bacterial phenotypic diversification—a potential bet-hedging strategy in chronic infections. Our previous work demonstrated that cells with higher endocytic capacity are more permissive to Mtb entry but also more efficient at delivering internalized bacteria to lysosomes (Sachdeva et al., 2020) - posing a potential conundrum for Mtb. The pathogen appears to face a trade-off between infecting cells that are easier to enter but potentially more hostile intracellularly, versus cells that are harder to access but offer a more permissive intracellular environment. The current findings reinforce this trade-off by confirming that Mtb residing in high endocytic capacity or G2 subpopulations experience higher oxidative stress. Interestingly, our identification of cell cycle progression as a key driver of endocytic heterogeneity, alongside prior evidence that Mtb lipids can arrest host cell cycle at the G1/S boundary (Cumming et al., 2017), offers a potential resolution. It is plausible that Mtb exploits host cell cycle dynamics to its advantage by gaining entry into subpopulations with high endocytic activity or in G2 phase, then modulating the host cell cycle to maintain a G1-like state more conducive to its survival. The partial and heterogeneous nature of this arrest could create a dynamic cellular landscape where subpopulations of bacteria experience different intracellular conditions, facilitating phenotypic diversification or temporal regulation of stress exposure. Longitudinal single-cell tracking of both host and bacterial states will be essential to test this model and clarify the temporal interplay between Mtb physiology, endocytic capacity, and cell cycle states. In the broader context of infection biology, this study underscores the importance of considering host cell state as an active participant in shaping infection trajectories and microbial fate decisions.

Several important questions arise from these findings. First, what are the molecular regulators that link cell cycle progression to changes in the endocytic machinery? Our data point to Rab5 and EEA1 abundance and vesicle dynamics as effectors, but the upstream regulatory nodes remain to be defined. Second, how generalizable is the principle that cell cycle–dependent endocytic variation drives bacterial phenotypic adaptation? It will be important to test whether other intracellular pathogens, such as Salmonella or Listeria, experience similar variation in intracellular fate depending on host cell cycle or trafficking capacity. Third, does tissue-level variation in the distribution of host cell states (e.g., cycling versus quiescent macrophages) create spatial patterns of bacterial adaptation in vivo? Fourth, dissecting whether therapeutic modulation of cell cycle or endocytic pathways can bias bacterial populations toward more drug-sensitive states presents an exciting translational opportunity.

A key implication of our work is that, even in genetically identical cell populations, temporal progression through the cell cycle generates predictable fluctuations in pathogen permissiveness. This offers a new lens to interpret single-cell data from infections: what appears as stochastic heterogeneity may in fact emerge from structured host programs. By connecting single-cell variation in a host trait to phenotypic diversity in an intracellular pathogen, we provide a conceptual framework for understanding infection as an interaction between two dynamic, heterogeneous cellular subpopulations. Individual bacteria within this framework are likely to exhibit substantial differences in their susceptibility to antibacterial drugs (Sherry et al., 2024; Dhar and J. D. McKinney, 2007; Dewachter et al., 2019; Cotten et al., 2023). Identifying and defining such hotspots of drug tolerance within this matrix could provide a means to specifically target recalcitrant bacterial subpopulations. Indeed, Mtb of different redox states are known to differ in drug susceptibility (Bhaskar et al., 2014). Our findings identify host endocytic heterogeneity as a tunable parameter that shapes bacterial phenotypic landscapes. Modulating this axis—either by synchronizing host states or targeting pathways coupling cell cycle and endocytosis—may provide a route to bias Mtb populations away from stress-tolerant phenotypes and enhance drug efficacy.

In conclusion, we establish a direct mechanistic link between host cell cycle progression and the phenotypic landscape of an intracellular pathogen. By demonstrating that predictable host rhythms can structure bacterial diversity, this work offers a new lens for interpreting heterogeneity in hostpathogen interactions and provides a foundation for mapping these complex dynamics in vivo.

Data availability

Single cell RNA seq raw files are submitted with GEO accession GSE300944.

Supplementary Figures

Figure S1:

Figure S2:

Figure S3:

Figure S4:

Figure S5:

Figure S6:

Figure S7:

Acknowledgements

We thank Amit Singh for the kind gift of the Mrx1-roGFP2 construct and, together with Dimple Notani, for critical reading of the manuscript. HeLa BAC GFP-Rab5 cells were kindly provided by Marino Zerial (MPI-CBG, Dresden). We also thank Vikas Yadav for assistance with Mtb redox data analysis and Prasanth Pillai for technical support. The following reagent was obtained through BEI Resources, NIAID, NIH: Macrophage Cell Line Derived from Wild Type Mice, NR-9456. We acknowledge the biosafety, central imaging and flow cytometry (CIFF), screening facilities, and the animal care and resource center (ACRC) at NCBS. The study was approved by the Institutional Animal Ethics and Biosafety Committees of NCBS. V.S. acknowledges funding from ANRF (CRG/2022/007379) and core support from NCBS-TIFR; N.S. acknowledges a fellowship from CSIR.

Additional files

Movie S1A

Movie S1B

Additional information

Funding

Department of Atomic Energy, Government of India (DAE) (RTI 4006)

• Varadharajan Sundaramurthy

DST | Science and Engineering Research Board (SERB) (CRG/2022/007379)

• Varadharajan Sundaramurthy