3.1 Tumor-initiating CSCs are positively correlated with immunosuppressive Treg cells in breast tumor tissue

To evaluate the relationship between breast CSCs and Tregs, if any, we undertook a multipronged approach to obtain cross-validating datasets. To that end, in our first approach, our in-silico study based on RNA-profiling dataset (GSE25066) available from ‘R2: Genomics Analysis and Visualization Platform’ using CSC (OCT4 and NANOG) as well as, Treg (FOXP3) signatures revealed a positive correlation (r = 0.1185, p < 0.01; OCT4 and FOXP3) (r = 0.1657, p < 0.001; NANOG and FOXP3) (n = 508) between breast CSCs and Tregs (Fig. 1A left panel). We extended our observation to include information regarding other tumor tissues as well by mining publicly available tumor datasets from ‘R2: Genomics Analysis and Visualization Platform’ and ‘The Cancer Genome Atlas’ (TCGA). Data obtained were in line with our findings in breast tumor tissue, i.e., CSC (ALDH1A1) and Tregs (FOXP3) markers possessed a positive correlation in glioma (dataset: Tumor Brain Lower Grade Glioma (2022-v2) - tcga − 532 - tpm - gencode36) (r = 0.1515, p < 0.001, n = 532) and prostate adenocarcinoma (dataset: Tumor Prostate Adenocarcinoma-TCGA-497- rsem–tcgars) (r = 0.1529, p < 0.001, n = 497) (Fig. 1A right panel).

In our second approach, Kaplan-Meier survival curve revealed lower probability of overall survival (OS) in breast tumor patients with higher CSC (NANOG) (p < 0.05) and Treg (FOXP3) (p < 0.05) signature genes (Fig. 1B), in concordance with earlier reports [36,37,38]. Third, to broaden the scope of our data, we performed immunohistochemistry (IHC) of breast tumor patient-derived tissues, which revealed an augmented incidence of stemness factors (OCT4, SOX2, NANOG) along with Treg signature gene FOXP3 (Fig. 1C) in high-grade (n = 4) than low-grade (n = 4) breast tumor tissues. Moreover, we observed and calculated the number of OCT4/SOX2/NANOG/FOXP3 positive cells per area taken from sections of the tumor tissues (Fig. 1C), and consequently found similar trend in stemness genes and FOXP3 coincides with high-grade of breast tumor compared to low-grade breast tumor (p < 0.0001) (Fig. 1D). Thus, raising the possibility of a direct relationship between them due to their increased levels in breast cancer.

In our fourth approach, we employed RNA expression datasets from ‘R2: Genomics Analysis and Visualization Platform’ database to identify the breast cancer subtype possessing highest percent of CSCs. Our analysis revealed significantly higher CSC-related stemness genes (OCT4, SOX2, NANOG) (p < 0.0001) and ALDH1A1 (p < 0.0001) in TNBC (GSE30682), in contrast to ER+ luminal breast cancer (GSE5460) (Fig. 1E). Reflecting these results, when we checked the CSC contents of different breast cancer cell lines, we obtained higher percent of (CD44+/CD24−) CSCs in TNBC cell lines, i.e., MDA-MB-468 and MDA-MB-231, as compared to luminal MCF-7 cell line (Fig. 1F). Among these two TNBC cell lines, MDA-MB-468 furnished the highest percentage of CSCs (~ 20%) (Fig. 1F). Concurrently, ‘R2: Genomics Analysis and Visualization Platform’ database analysis demonstrated significantly higher FOXP3 expression (p < 0.0001) in TNBC (GSE30682), when compared with ER+ luminal subtype (GSE5460) (Fig. 1E, right panel) validating our results. To understand whether the mutuality between CSCs and Treg is reflected in TNBCs as well, we performed RNA profiling dataset analysis from ‘R2: Genomics Analysis and Visualization Platform’ database and observed a positive correlation between ALDH1A1 and FOXP3 in TNBC patient dataset (GSE76714) (p < 0.0001, r = 0.4907, n = 71) (Fig. 1G).

Since, the expressions of the gene signatures do not necessarily correlate with the actual number of cells, in our fifth approach, we directly evaluated the percentage of CD4+CD25+FOXP3+ Tregs from breast tumor patient-derived blood samples. In line with our above findings, this experiment also revealed significantly higher Treg (p < 0.01) percentages in high-CSC-containing (Fig. 1E) TNBC patients’ (n = 4) blood as compared to non-TNBC patient (n = 4) cohort (Fig. 1H).

Such step-wise multi-approach investigations involving in silico and patient tissue analysis strongly indicated a direct relation between CSCs with Tregs.

3.2 CSCs, even in low numbers, are able to generate Treg cells

Above results indicating a direct relationship between CSCs and Tregs in breast tumor tempted us to explore as to how during tumor initiation, CSCs, although present in low number [39] in comparison to surrounding anti-tumor effector CD4+ T cells, not only evade immune-elimination but also proliferate and form the entire tumor mass by generating NSCCs. To that end, human TNBC cell line MDA-MB-468 was utilized for generating CSC-enriched secondary mammospheres (Fig. 2A) since these breast cancer cells furnished highest CSC percentage (Fig. 1F). MDA-MB-468-derived secondary mammospheres generated ~ 3.5-fold CD44+/CD24− CSC enrichment as compared to MDA-MB-468 cells (p < 0.0001) (Online Resource 3, Supplementary Fig. 1A). From these secondary mammospheres, breast CSCs were purified using Miltenyi bead isolation protocol (Schematic diagram depicted in Online Resource 3, Supplementary Fig. 1B). Flow cytometry data further confirmed > 90% purity of CD44+/CD24− CSCs (Online Resource 3, Supplementary Fig. 1C). These CSCs were further characterized to obtain a significant higher expression of (i) stemness markers OCT4 (p < 0.001), SOX2 (p < 0.01), and NANOG (p < 0.01); (ii) EMT markers N-CADHERIN (p < 0.01), and VIMENTIN (p < 0.01), as well as (iii) drug resistance markers ABCG2 (p < 0.01), MRP1(p < 0.01), and MDR1 (p < 0.01) (Online Resource 3, Supplementary Fig. 1D), when compared with corresponding NSCCs (CD44−/CD24−, CD44−/CD24+, CD44+/CD24+). These results together establish that CD44+/CD24− breast cancer cell subpopulation was indeed CSC in nature.

Fig. 1

Tumor-initiating CSCs and immunosuppressive Treg cells have positive correlation in breast cancer. A Plots showing correlation between NANOG and FOXP3, OCT4 and FOXP3, from GSE25066 breast cancer patient dataset (left). Plots demonstrating positive correlation between cancer stemness marker ALDH1A1 and suppressive Treg signature FOXP3 in brain cancer glioma (dataset used: Tumor Brain Lower Grade Glioma (2022-v2)-tcga-532-tpm-gencode36) (middle) and prostate cancer (dataset used: Tumor Prostate Adenocarcinoma-TCGA-497-rsem-tcgars) (right). Datasets were obtained from ‘R2: Genomics Analysis and Visualization Platform’ and ‘TCGA’ database. “r” is correlation co-efficient. B Kaplan-Meier (KM) plot depicting low overall survival (OS) probability of breast tumor patients harboring high expression of CSC marker NANOG and Treg signature gene FOXP3. C Representative Immunohistochemistry (IHC) images showing elevated expression of stemness markers OCT4, SOX2, NANOG, and Treg signature gene FOXP3 in low-grade vs. high-grade breast tumor tissues (n = 4 in each group). Scale bar = 50 µM and magnification 40X. D Bar diagram showing number of OCT4, SOX2, NANOG, and FOXP3 positive cells in low-grade vs. high-grade breast tumor tissues as determined by IHC-staining. E Bar graphs showing incidence of higher cancer stemness markers OCT4, SOX2, NANOG, and ALDH1A1 along with suppressive Treg marker FOXP3 in TNBC tissues than ER+-luminal breast tumor tissues. Bar graphs were plotted using RNA profiling data of TNBC dataset (GSE30682) and ER+-luminal breast tumor dataset (GSE5460) available in ‘R2: Genomics Analysis and Visualization Platform’. F Flow-cytometry data showing percent CD44+CD24− CSC population in ER+-luminal MCF-7, triple-negative MDA-MB-231, and MDA-MB-468 breast cancer cells. G Plot showing a positive correlation between ALDH1A1 and FOXP3 in TNBC dataset (GSE76714) available in ‘R2: Genomics Analysis and Visualization Platform’. H Bar graph demonstrating occurrence of higher Treg cell percentage in TNBC patients than non-TNBC patients. Data were represented as the mean ± SD of minimum 3 independent experiments performed in triplicate. Student’s t-test (unpaired) (D, E, H) and one-way ANOVA (F) was used to assess the data where *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. BC: breast cancer, TNBC: triple-negative breast cancer

Considering the scenario that during initiation only a low number of CSCs remain in the site of tumor origin in comparison to higher number of infiltrating T cells [40, 41], next in an attempt to mimic such condition, anti-CD3/anti-CD28-activated CD4+ T cells (for 72 h) were cultured in the CM of 72 h culture of purified MDA-MB-468 CSCs, at the CSC:T cell ratio of 1:5, as depicted by the schematic diagram (Fig. 2B). Our flow-cytometry results (Fig. 2C, left panel) show that CM from even such low number of CSCs was able to generate significantly higher CD4+CD25+FOXP3+ Treg cells (p < 0.01) as compared to activated T cell counterpart (treated with anti-CD3/anti-CD28) (Fig. 2C, right panel). Activated T cells treated with IL2 and TGFβ1 to generate Treg were taken as positive control (Fig. 2C) [27, 42].

To evaluate the immunosuppressive properties of these newly generated Treg cells, we magnetically sorted CD4+CD25+CD127− Treg cells (Online Resource 3, Supplementary Fig. 1E) since CD4+CD25+CD127− Treg cells mostly correspond to CD4+CD25+FOXP3+ Treg cell population [43,44,45]. To that end, CD4+ T cells were labelled with CFSE and co-cultured those with CSC CM-induced Treg cells. Results of Fig. 2D demonstrated significant inhibition (p < 0.01) in proliferation of the activated CD4+ T lymphocytes in presence of the CSC-induced Treg cells. Furthermore, upon co-culturing with Treg fraction, CD4+ effector T cells showed significant decrease (p < 0.01) in IFNγ expression (Fig. 2E), which is a signature secretome of effector T cells [46].

All these findings together confirmed that tumor-initiating CSCs, despite of being present in low numbers as compared to the effector T cells, transformed the latter into immunosuppressive Treg cells in contact-independent manner to ensure escape from immune-elimination.

3.3 Chemotherapy-spared CSCs induce Treg generation mimicking the condition of tumor initiation during relapse

Plethora of reports demonstrate the failure of conventional chemotherapy to kill CSCs [8, 47, 48]. Reports from our lab as well as others have further established the enrichment of CSC repertoire during chemotherapy [8, 49]. Furthermore, an enrichment of RNA transcripts of CSC-associated markers upon chemo-treatment was also noticed in TNBC that ultimately resulted in poor relapse-free survival (RFS) [50, 51]. In this relation, our Kaplan-Meier analyses revealed that NACT-treated breast cancer patients with elevated OCT4 and NANOG have significantly lower RFS (OCT4, p < 0.05; NANOG, p < 0.01) (Online Resource 4, Supplementary Fig. 2A and B) in comparison to patients furnishing lower levels of OCT4 and NANOG. These results as well as all above-mentioned reports tempted us to explore whether these chemo-escaped CSCs further generate immune-suppressed microenvironment thus subsequently causing tumor relapse after withdrawal of the treatment. For the same, we again undertook multi-approach experimentations as described below.

Fig. 2

Low number of CSCs are sufficient to generate immunosuppressive Treg cells. A Inverted microscopic image showing the appearance of monolayer MDA-MB-468 cells and cell-derived spheroids viewed under 10X magnification. B Schematic representation of immunosuppressive Treg cell generation using CSC-CM taking single-cultured CSCs and T cells in 1:5 ratio. C FACS plots (left panel) and representative graph (right panel) showing immunosuppressive Treg cell percentage in (i) only α-CD3 + α-CD28, (ii) α-CD3 + α-CD28 + TGFβ + IL2, and (iii) α-CD3 + α-CD28 + CSC-CM. D Flow-cytometric histoplots (left panel) and representative bar graph (right panel) showing T-cell proliferation in presence of CSC-induced Treg cells. E FACS plots (left panel) and bar graph (right panel) furnishing percentage of IFNγ secreting CD4+ T cells after co-culture with CSC-induced Treg cells. Data were represented as the mean ± SD of minimum 3 independent experiments performed in triplicate. Student’s t-test (unpaired) was used to assess the data where *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. CSCs: cancer stem cells; CM: conditioned medium

First, MDA-MB-468 cells were treated with doxorubicin (2.5 µM) for 24 h, that induced significant apoptosis in NSCC fraction (p < 0.0001) but failed to do so in CSCs (Fig. 3A). At this juncture, we considered the following information: (i) after multiple rounds of chemo-treatment, NSCCs are targeted and killed causing a decrease in tumor size while rare CSCs survive these therapeutic regimens, and ‘recreate’ the tumor after anti-cancer therapy [52, 53], either in the same or secondary organ; and (ii) increase in stemness markers depicting poor RFS (Online Resource 4, Supplementary Fig. 2A and B). These led us to hypothesize that such drug-surviving CSCs might be causing relapse after chemotherapy. Consequently, in our second approach, we treated MDA-MB-468 primary spheres (containing ~ 30% CSCs) with doxorubicin (2.5 µM) for 24 h then the treatment was removed and spheres were incubated in fresh media for another 48 h; this sequence was carried out for 3 cycles to mimic multiple chemotherapy sessions that may lead to relapse later (Schematic diagram Fig. 3B). Our results showed that NSCCs underwent significant apoptosis with each cycle (cycle 1 p < 0.001, cycle 2 p < 0.0001, cycle 3 p < 0.05) (Fig. 3C), whereas, CSCs could escape the treatment even after the last cycle of chemotherapy (cycle 3) since no apoptosis was noticed in this CSC fraction after each cycle (Fig. 3C). Our representative graph depicting cell number pre- and post- 3 cycles-treatment (Fig. 3D right panel), showed that while ~ 100,000 cells (30% CSCs and 70% NSCCs as seen in Fig. 3D left panel) were present before treatment, after the third cycle of treatment ~ 57,000 cells were alive. Among these chemo-escaped cells, 90% was CSCs and 10% NSCC (Fig. 3D left panel), indicating that the reduction in total cell number was due to killing of NSCCs, and majority of drug-surviving cells were CSCs. We next investigated the status of Treg cell generation after chemotherapy. To that end, in our third approach, in-silico Kaplan-Meier data of NACT-treated breast cancer patients showed positive correlation between higher FOXP3 expression (p < 0.05) (Online Resource 4, Supplementary Fig. 2C) with lower RFS, as compared to NACT patients expressing lower FOXP3, reflecting the correlation pattern we found with higher stemness (OCT4, NANOG) and lower RFS (Online Resource 4, Supplementary Fig. 2A and B) under similar conditions. Fourth, immunohistochemical analysis of breast cancer patient tissue samples revealed higher expression of CSC-related stemness genes (OCT4, SOX2, and NANOG) along with Treg marker FOXP3, in chemotherapy-treated cohort (n = 5), as compared to non-chemotherapy treated group (n = 5) (Online Resource 4, Supplementary Fig. 2D).

To validate the same, in our fifth approach, we isolated CSCs from untreated primary spheres and from the surviving population after 3rd cycle of chemotherapy. Interestingly, we observed fold increase in mean fluorescence intensity (MFI) of stemness factors Oct4 (p < 0.01), Sox2 (p < 0.0001), and Nanog (p < 0.0001) after the 3rd cycle chemotherapy-escaped CSC in contrast to untreated primary sphere-derived CSCs (Online Resource 4, Supplementary Fig. 2E), indicating an increase in stemness per CSC after 3 consecutive rounds of doxorubicin treatment. Next, anti-CD3/anti-CD28-activated CD4+ T cells were cultured in the CM collected from same numbers of CSCs from both the sets in a CSC: T cell ratio of 1:5, for another 72 h (Schematic diagram shown in Fig. 3E). Our results depicted significant immunosuppressive Treg generation (p < 0.001) from CM of CSCs isolated post-3 cycles of chemotherapy as compared to CSCs isolated from untreated MDA-MB-468-derived primary mammospheres (Fig. 3F).

Above-mentioned multi-pronged study establishes generation of immunosuppressive Treg cells by chemo-escaped CSCs during condition mimicking tumor relapse. Since chemotherapy itself causes immunosuppression [54], remaining T cells might be getting converted to Treg cells by drug-spared CSCs thereby aggravating that situation and ensuring survival of the drug-surviving CSCs for initiating tumor again thereby, causing tumor relapse after chemotherapy.

3.4 CSCs generate Treg cells by secreting TGFβ

Next, we aimed at deciphering the mechanism underlying CSC-induced Treg cell generation. To that end our literature search revealed the crucial role of the cytokine TGFβ in Treg cell generation [55,56,57]. Further reports also specified the release of TGFβ by cancer cells [58, 59]. In line with this, murine pancreatic cancer cells reportedly converted naïve T cells to CD4+FOXP3+ Treg cells by releasing TGFβ [60]. This information together led us to hypothesize whether CSC-shed TGFβ is the molecule behind such generation of Treg cells during tumorigenesis.

To that end, supporting our hypothesis, in-silico RNA profiling database from ‘R2: Genomics Analysis and Visualization Platform’ showed a positive correlation between CSC marker ALDH1A1 and TGFβ1 (GSE 69,031) (r = 0.1863; p < 0.05) (Fig. 4A left panel) as well as Treg signature FOXP3 and TGFβ1 (GSE5460) (r = 0.2955; p < 0.001) (Fig. 4A right panel) in breast cancer patients. Again, ‘R2: Genomics Analysis and Visualization Platform’ database analysis revealed a positive correlation of ALDH1A1 with TGFβ (r = 0.2126, p < 0.01, n = 226) and FOXP3 with TGFβ (r = 0.2697, p < 0.0001, n = 226) in TNBC subset (GSE142102) as well (Fig. 4B).

Furthermore, according to Shipitsin et al. TGFβ is up-regulated in CD44+-expressing breast cancer cells [61]. Our ELISA results depicted that TGFβ level was significantly higher in MDA-MB-468 CSC-CM than NSCC-CM (p < 0.001) (Fig. 4C). Next, anti-CD3/anti-CD28-treated T cells were exposed to (i) CM of Miltenyi bead purified CSC (CSC: T cell ratio 1:5) or (ii) stimulation by IL2 and TGFβ [27]. We observed that, CSC-CM could generate ~ 15% Treg cells (which was comparable to that formed (~ 18%) upon stimulation by IL2 and TGFβ, taken as positive control) in comparison to anti-CD3/anti-CD28-treated CD4+ T cells (Fig. 4D).

Fig. 3

Doxorubicin-spared CSCs induce immunosuppressive Treg cells mimicking tumor-initiation following tumor relapse. A Bar graph showing percentage of apoptotic NSCCs and CSCs populations gated from MDA-MB-468 cells after doxorubicin treatment, as determined using flow-cytometry. B Schematic representation of three cycles of doxorubicin treatment regimen to MDA-MB-468 cell-derived spheroids. C Bar diagrams depicting percentage of apoptotic NSCCs and CSCs after 1st cycle (left), 2nd cycle (middle), and 3rd cycle (right) of chemotherapy. Percent apoptosis was flow-cytometrically determined using Annexin-V binding assay. D Representative bar diagram showing percentage of CD44+/CD24− CSCs in control vs. post- 3 cycle doxorubicin-treated MDA-MB-468-derived spheroids (left). Bar plot showing total number of cells counted using hemocytometer from control vs. post-3 cycledoxorubicin-treated spheroids. E Schematic representation of Treg cell generation using MDA-MB-468 spheroid-derived CSC-CM and CSC-CM after 3 cycles of doxorubicin treatment. F Flow-cytometric plots showing percentage of Treg cells (left) and bar graph (right panel) showing percentage of suppressive Treg cells generated from activated T cells in presence of doxorubicin-treated or untreated MDA-MB-468 spheroid-derived CSC-CM. Data were represented as the mean ± SD of minimum 3 independent experiments performed in triplicate. Student’s t-test (unpaired) was used to assess the data where *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. CSCs: cancer stem cells; Sph: spheroids; CM: conditioned medium

Since we have confirmed earlier that our CSC-CM is enriched in TGFβ level (Fig. 4C), this observation directed us to hypothesize that CSC-shed TGFβ might be involved in the augmented generation of Treg cells. However, possibility of the contribution of IL2, if at all secreted by CSCs, has not been negated. Finally, to re-affirm the contribution of CSC-shed TGFβ in Treg formation, when we neutralized TGFβ in CSC-CM, a significant decrease (p < 0.01) in Treg percentage, validating the essential role of CSC-released TGFβ in Treg cell generation (CSC: T cell ratio 1:5) (Fig. 4E).

3.5 Chemotherapy further increases TGFβ in CSCs thus resulting in a higher Treg polarization favoring tumor relapse

Above results together tempted us to hypothesize that chemotherapy might further increase TGFβ in CSCs thereby resulting in a higher Treg polarization and culminating in a microenvironment that favors tumor relapse. To validate our hypothesis, TGFβ expression was evaluated flow-cytometrically in the isolated CD44+/CD24− CSC population in presence of PMA + Ionomycin + Brefeldin A (last 5 h of culture) with or without multiple rounds of doxorubicin treatment. A significant (p < 0.001) increase in TGFβ content (~ 2.2 fold) after the third cycle of doxorubicin-treated MDA-MB-468 CSCs was obtained in comparison to untreated CSCs (Fig. 4F). To revalidate these results, we next checked secretory TGFβ levels in the CM of above-mentioned cells by ELISA and found that 3 cycles of chemotherapy significantly (p < 0.0001) increased CSC-shed TGFβ in the CM than the same collected from non-treated CSCs (Fig. 4G).

These results signify that chemotherapy-escaped CSCs, by secreting more TGFβ, generate more Treg cells than the untreated tumor-initiating CSCs, thereby facilitating immune-evasion and ensuring recurrence even after a few months to several years [9].

This shows how CSCs even present in low numbers during tumor initiating or relapse phase, are able to evade immune destruction by converting CD4+ T cells to immunosuppressive Treg cells. It would not be out of context to mention at this point that at tumor initiating conditions, expression of IFNγ, a key Th1 response cytokine [62] was suppressed when anti-CD3/anti-CD28-treated CD4+ T cells were exposed to CM of MDA-MB-468 CSC (CSC: T cell ratio 1:5) in comparison to unexposed anti-CD3/anti-CD28-treated CD4+ T cells (p < 0.01) (Online Resource 5, Supplementary Fig. 3). However, as compared to control CSC-CM, when TGFβ was neutralized in CSC-CM, an increase in IFNγ expression was noted in the activated T cell subset (p < 0.05), validating the role of CSC-shed TGFβ in decreasing Th1 response. Additionally, recombinant TGFβ, when directly added to activated CD4+ T cells, decreased IFNγ levels (p < 0.001) (Online Resource 5, Supplementary Fig. 3). These findings signify the contribution of CSC-shed TGFβ in creating an immunosuppressive environment in more ways than one, to ensure proper tumor development.

Fig. 4

CSC-secreted TGFβ induces immunosuppressive Treg cell generation. A Plot showing positive correlation between CSC markers ALDH1A1 and TGFβ1, from GSE69031 breast cancer patient dataset (left) and representative plot demonstrating correlation between TGFβ1 and suppressive Treg marker FOXP3 using GSE5460 breast cancer dataset (right panel). Datasets were obtained from ‘R2: Genomics Analysis and Visualization Platform’ database. “r” is correlation co-efficient. B Plots showing a positive correlation of ALDH1A1 with TGFβ (left panel) and FOXP3 with TGFβ (right panel) in TNBC dataset (GSE142102) available in ‘R2: Genomics Analysis and Visualization Platform’. C Representative bar plots showing level of secreted TGFβ in the CM of MDA-MB-468-derived CSCs and NSCCs, as determined by ELISA. D FACS plots (left panel) and bar diagram (right panel) depicting percent immune-suppressive Treg cells generated from CD4+ T cells following activation by (i) only α-CD3 + α-CD28, (ii) α-CD3 + α-CD28 + TGFβ + IL2, and (iii) α-CD3 + α-CD28 + CSC-CM. E FACS plots (left panel) and bar diagram (right panel) depicting percent Treg cell generation from activated T cells cultured in presence of CSC-CM pre-incubated with or without anti-TGFβ antibody. F Bar plot demonstrating expression of TGFβ per-CSC in control vs. 3-cycles of doxorubicin-treated CSCs derived from MDA-MB-468 spheroids. G Representative bar plot showing the amount of TGFβ present in the CM of control vs. 3-cycles of doxorubicin-treated CSCs derived from MDA-MB-468 spheroids determined by ELISA. Data were represented as the mean ± SD of minimum 3 independent experiments performed in triplicate. Student’s t-test (unpaired) was used to assess the data where *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. CSC: cancer stem cell; MFI: mean fluorescence intensity; CM: conditioned medium