Three-dimensional analysis of mitochondria in a patient-derived xenograft model of triple negative breast cancer reveals mitochondrial network remodeling following chemotherapy treatments

Abstract

Mitochondria are hubs of metabolism and signaling and play an important role in tumorigenesis, therapeutic resistance, and metastasis in many cancer types. Various laboratory models of cancer demonstrate the extraordinary dynamics of mitochondrial structure, but little is known about the role of mitochondrial structure in resistance to anticancer therapy. We previously demonstrated the importance of mitochondrial structure and oxidative phosphorylation in the survival of chemotherapy-refractory triple negative breast cancer (TNBC) cells. As TNBC is a highly aggressive breast cancer subtype with few targeted therapy options, conventional chemotherapies remain the backbone of early TNBC treatment. Unfortunately, approximately 45% of TNBC patients retain substantial residual tumor burden following chemotherapy, associated with abysmal prognoses. Using an orthotopic patient-derived xenograft mouse model of human TNBC, we compared mitochondrial structures between treatment-naïve tumors and residual tumors after conventional chemotherapeutics were administered singly or in combination. We reconstructed 1,750 mitochondria in three dimensions from serial block-face scanning electron micrographs, providing unprecedented insights into the complexity and intra-tumoral heterogeneity of mitochondria in TNBC. Following exposure to carboplatin or docetaxel given individually, residual tumor mitochondria exhibited significant increases in mitochondrial complexity index, area, volume, perimeter, width, and length relative to treatment-naïve tumor mitochondria. In contrast, residual tumors exposed to those chemotherapies given in combination exhibited diminished mitochondrial structure changes. Further, we document extensive intra-tumoral heterogeneity of mitochondrial structure, especially prior to chemotherapeutic exposure. These results highlight the potential for structure-based monitoring of chemotherapeutic responses and reveal potential molecular mechanisms that underlie chemotherapeutic resistance in TNBC.

INTRODUCTION

Triple negative breast cancer (TNBC), an aggressive subtype of breast cancer, comprises 10–15% of breast cancer cases¹. Because TNBCs do not express estrogen or progesterone receptors or overexpress the human epidermal growth factor receptor 2, there are limited targeted therapies for this subtype^1,2. Most patients with TNBC receive pre-surgical neoadjuvant chemotherapy (NACT) that includes several conventional chemotherapeutic agents, usually given in simultaneous and/or sequential combinations. NACT now also includes the immunotherapy agent Programmed Cell Death 1 (PD-1) inhibitor with a marginal but statistically significant improvement in response rate³. Conventional chemotherapies typically include DNA-damaging agents (e.g., Adriamycin, cyclophosphamide, carboplatin) and taxanes (e.g., docetaxel, paclitaxel). As there is no single accepted standard NACT regimen for TNBC, various regimens are used throughout the United States⁴. Residual cancer burden (RCB) following NACT is assessed by a pathology evaluation of the resected surgical specimen. Unfortunately, approximately 45% of TNBC patients retain substantial RCB after NACT, which is strongly associated with high rates of metastatic relapse and poor overall survival^5,6. Thus, understanding the molecular features of residual tumor cells that survive chemotherapeutic regimens will be key to developing novel strategies to combat residual disease.

Mitochondria are adaptive organelles that allow cells to meet energetic and signaling demands in response to exogenous stressors, in part through mitochondrial structural rearrangements⁷. In the context of cancer, metabolic adaptations can encompass oxidative phosphorylation (oxphos), glycolysis, reactive oxygen species (ROS), nucleotide synthesis, and fatty acid oxidation as we recently reviewed⁸. The interplay of mitochondrial fission and fusion that regulates mitochondrial structure has been studied in cancer therapy resistance and metastasis. For TNBC, mitochondrial structure dynamics play a role in metastatic phenotypes, although the findings seem to be context-dependent^9,10. We showed that mitochondrial fusion drove to chemotherapy-induced oxphos that promoted the survival of residual tumor cells that were refractory to conventional chemotherapies¹¹. Importantly, we demonstrated that chemotherapy-induced mitochondrial structure and function adaptations resulted in a new therapeutic dependency for residual TNBC, wherein we could delay tumor regrowth by inhibiting mitochondrial fusion or by direct inhibition of Complex I^11–13.

Serial block-face scanning electron microscopy (SBF-SEM) was first developed in 2004 by Denk and Horstmann to study mitochondrial structure and location within the neuron^14,15. SBF-SEM produces high-resolution micrographs that capture subcellular ultrastructural features through three-dimensional (3D) rendering^16,17 that, unlike other volumetric electron microscopy techniques, provide larger x- and y-axis ranges^16,17. These visualizations of changes in mitochondrial morphology, orientation, configuration, size, and shape provide insights into mitochondrial physiology and function important for physiology and disease^18–21. We and others have shown that using 2D transmission electron microscopy (TEM) to characterize mitochondrial structures in breast cancer provides useful, albeit limited, insights into the extent of the mitochondrial network^10,11,22,23. Previous SBF-SEM studies of mitochondria demonstrate dynamic 3D structures and networks associated with the physiology of skeletal muscle²⁴, aging^25–27, pancreatic ductal adenocarcinoma²⁸, and non-small cell lung cancer (NSCLC)²⁹. Of note, the recent NSCLC study employed genetically engineered mouse models to demonstrate that 3D mitochondrial structure was spatially heterogeneous within tumor cells and functionally affected mitochondrial respiration²⁹.

In TNBC, the comprehensive nature of 3D mitochondrial networks remains unexplored. Further, SBF-SEM has not been applied to patient-derived xenograft (PDX) models nor to understand therapeutic resistance in cancer. To address these knowledge gaps and based on our prior work demonstrating the importance of mitochondrial structure and oxphos for TNBC chemoresistance, we used SBF-SEM to characterize 3D mitochondrial structures in our deeply characterized orthotopic PDX model of TNBC. PDXs provide valuable models for longitudinal studies to evaluate chemotherapy response as they effectively recapitulate the molecular features and the heterogeneity observed in human TNBC^30–34. Our longitudinal analyses enabled us to compare residual tumors surviving exposure to a variety of conventional chemotherapies to the treatment-naïve state. We found significant changes in mitochondrial shape, size, and complexity in residual tumors compared to the treatment-naïve tumor. Of note, these changes were more pronounced in tumors that had been exposed to single, rather than combination, chemotherapies. Further, we found striking intratumoral heterogeneity (ITH) of mitochondria within individual PDX tumors, and this was most pronounced in the treatment-naïve tumor. This finding highlights the incredible complexity within individual TNBC tumors and reaffirms the extensive ITH that has been thoroughly documented at the genomic, transcriptomic, epigenetic, and proteomic levels^35,36.

METHODS

Patient-derived xenograft mouse studies

Animal studies were conducted in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals with approval of the BCM IACUC under protocol AN-8243. Mice were ethically euthanized as recommended by the Association for Assessment and Accreditation of Laboratory Animal Care. The PIM001-P PDX model was originally obtained from the University of Texas MD Anderson Cancer Center, where it was generated and characterized and was then propagated as described previously^13,32. Briefly, cryo-preserved PDX cells were suspended as a 1:1 volume mixture in a total of 20 μl of complete cell growth medium (Dulbecco’s modified Eagle’s medium:F12, Cytiva HyClone, SH30023.01) supplemented with 5% fetal bovine serum and Matrigel (Corning, 354234). 5.0 × 10⁵ viable tumor cells were injected unilaterally into the fourth mammary fat pads of 5- to 8-week-old female Nod-Rag-Gamma mice (NRG, Nod.Cg-Rag1^tm1Mom/ IL2rg^tm1wjl/SzJ, The Jackson Laboratory).

To validate the PDX model’s identity, the Cytogenetics and Cell Authentication Core at MD Anderson Cancer Center performed short-tandem repeat (STR) DNA fingerprinting on DNA extracted from PDX tumors. The Promega 16 High Sensitivity STR Kit (Catalog # DC2100) was used for fingerprinting analysis, and profiles were compared to online databases (DSMZ/ ATCC/ JCRB/ RIKEN).

Mouse body weights and tumor volumes, measured using digital calipers, were recorded twice weekly for the duration of the experiment. When mammary tumors reached an average size of 150 mm³, mice were randomized into five treatment groups (n = 6 mice per group) to monitor tumor responses and collect samples for molecular analyses. For the treatment-naïve group, the tumors were harvested at 150 mm³ without any treatment. Mice in the Adriamycin plus cyclophosphamide (AC) treatment group received one dose of Adriamycin and cyclophosphamide every three weeks for a total of three doses. Adriamycin (doxorubicin, ChemieTek, CT-DOXO) was solubilized in sterile water immediately before administration and protected from light prior to intraperitoneal (i.p.) injection at a dose of 0.5 mg/kg. Cyclophosphamide (Sandoz, 0781-3244-94) was solubilized in sterile water immediately prior to administration via i.p. injection at 50 mg/kg. Mice in the docetaxel (DTX), the carboplatin (CRB), and the docetaxel plus carboplatin combination (DTX+CRB) treatment groups received weekly doses for three weeks. Docetaxel (Hospira, 0409-0201-10) was supplied in a solution of polysorbate 80NF, 4 mg anhydrous citric acid USP, 23% v/v dehydrated alcohol USP, q.s. with polyethylene glycol 300 NF at 20 mg/kg. DTX was administrated by i.p. injection at a dose volume of 2 ml/kg. Carboplatin (TEVA Pharmaceuticals, 0703-4248-01) was supplied at 10 mg/ml in water for i.p. injection at 50 mg/kg at a dose volume of 5 ml/kg. For the DTX+CRB treatment group, each compound was administered consecutively by i.p. injection. For each group, residual tumors were harvested one full cycle after the third dose of chemotherapy (i.e., after a total of 28 days for DTX, CRB, DTX+CRB, and after 63 days for AC.

Formalin-fixed paraffin-embedded (FFPE) tissue staining

After harvesting the tumors, we fixed tumor fragments in formalin (Sigma HT501128–4L) for 48 hours at room temperature on a rocker, washed them with 1× phosphate-buffered saline, and stored them in 70% ethanol at 4 °C until they were submitted to the BCM Pathology Core and Lab to be processed into paraffin blocks, cut into 3um sections, and stained with hematoxylin and eosin (H&E). Immunohistochemical staining with antibodies against Ku80 (CST 2180, 1:100) and human mitochondria (Abcam 92824, 1:1,000) was conducted as described previously¹³.

SBF-SEM sample processing

Tumors were harvested from the mammary gland immediately after mice were euthanized. A tumor slice ~1 mm thick was placed into 1 ml of fixative (2% paraformaldehyde + 2% glutaraldehyde in 0.15 M cacodylate buffer) and stored at 4°C until they were sent to the Mayo Microscopy and Cell Analysis Core for processing and imaging. Tissue samples for SBF-SEM were prepared using a modified procedure^20,37. Briefly, fixed tissue was rinsed in 0.1 M cacodylate buffer and placed in 2% osmium tetroxide + 1.5% potassium ferracyanide in 0.1 M cacodylate, washed with nH₂O (MilliQ water), incubated at 50°C in 1% thiocarbohydrazide, incubated again in 2% osmium tetroxide in nH₂O, rinsed in nH₂O and placed in 2% uranyl acetate overnight. The next day, the tissue was rinsed again in nH₂O, incubated with Walton’s lead aspartate, dehydrated through an ethanol series, and embedded in Embed 812 resin. To prepare embedded tissue for placement into the SBF-SEM, a ~1.0 mm³ piece was trimmed of excess resin and mounted on an 8 mm aluminum stub using silver epoxy Epo-Tek (EMS, Hatfield, PA). The mounted sample was then carefully trimmed into a smaller ~0.5 mm³ tower using a Diatome diamond trimming tool (EMS, Hatfield, PA) and vacuum sputter-coated with gold-palladium to help dissipate any charge. Sectioning and imaging of the sample were performed using a VolumeScope 2 SEM^™ (Thermo Fisher Scientific, Waltham, MA). Imaging was performed under low vacuum/water vapor conditions with a starting energy of 3.0 keV and a beam current of 0.10 nA. Sections of 50 nm thickness were cut, providing a final image spatial resolution of either 15 nm × 15 nm × 50 nm or 10 nm × 10 nm × 50 nm.

SBF-SEM data analysis

Image analyses including registration, volume rendering, and visualization, were performed using Amira (Thermo Fisher) software packages. The selection criteria for a region of interest (ROI) included tumor regions and excluded necrotic and extracellular matrix regions. A minimum of three ROIs were selected per sample. For each sample, images were collected from at least 300 serial sections that were then stacked, aligned, and visualized using Amira to make videos and quantify volumetric structures. A random number generator was used to select 350 segmented mitochondria for each tumor for the final analysis. See Supplementary Table 1 for the specific number of mitochondria segmented per ROI and those selected for final analysis. Previously defined protocols using Amira software provide a workflow for 3D reconstructed organelles by manual or semi-automated segmentation¹⁷. Amira provides parameters for quantifying 3D structures, many of which are included in the software, such as volume and surface area. We have added sphericity, which is measured through the Feret’s Diameter. Feret’s diameter refers to the distance between any two most distant points in a 3D object. Sphericity refers to the adherence of the object to that of a sphere and is calculated by the equation (π^1/3(6V)^2/3)/SA, where V is the volume of the object, and SA is the surface area of the object. We also quantified the mitochondrial complexity index, calculated by the equation SA³/16π²V², and the mitochondrial branching index, which calculates the relative branching between the transverse and longitudinal mitochondrial orientations²⁴.

Statistical analyses

We evaluated the means for each measurement between treatment groups by the Mann–Whitney U test followed by the Holm method of p-value adjustment^38–40. We performed Bartlett’s test followed by pairwise F-tests to determine whether the variance of each measurement differed between treatment groups using R 4.1.0 (R Core Team, 2023) and the stats package (base statistical package in R)⁴¹. Bartlett’s test assesses whether all the treatment groups have the same variance. A p-value less than 0.05 for a given measurement by Bartlett’s test indicates that some of the treatment groups have different variances for that measurement. To identify which treatment groups differed, we carried out a two-tailed F test on each possible pairwise comparison between treatment groups. In a two-tailed F test, the F value is defined as the quotient of the variance of one group divided by the variance of the other. This F value is used to test whether the variance in the numerator equals that in the denominator. For a given F test, a p-value was determined by taking the area under the F distribution beyond that test’s F value and then subtracting that area from 1. Once all possible comparisons had been made for a measurement, the p-values for that measurement were adjusted by the Holm method to account for multiple comparisons.

RESULTS

Residual tumors persisted in a PDX model following various chemotherapy treatments.

We previously deeply an orthotopic PDX model of primary treatment naïve TNBC, PIM001-P^13,32. Our previous work with the conventional front-line chemotherapy combination Adriamycin + cyclophosphamide (AC) revealed partial tumor regression followed by regrowth upon discontinuation of treatment. The regressed, aka ‘residual’, tumors had heightened oxphos compared to treatment-naïve tumors, resulting in a novel therapeutic window for inhibition of electron transport chain complex I¹³. We went on to show that following chemotherapy exposure, residual tumor cells had altered mitochondrial structure, which elicited heightened vulnerability to inhibition of mitochondrial fusion protein optic atrophy 1 (OPA1) with the small molecule inhibitor MYLS22^11,12.

Here, using the PIM001-P PDX model, we compared AC with docetaxel (DTX), carboplatin (CRB), or combined DTX+CRB (Fig. 1a). As we observed previously with AC treatment of PIM001-P¹³, tumors in all treatment groups exhibited partial regression. After three cycles of chemotherapy, residual tumors that were harvested (on the day that the fourth cycle would have began) showed an average decrease in tumor volume across all groups of 53.6% (DTX, 52.9%; CRB, 61.0%; DTX+CRB, 61.5%; and AC, 34.0%) (Fig. 1c). Animals maintained body weight with all chemotherapy regimens as we observed previously^13,42. For feasibility, we randomly selected one mouse per group to be analyzed by SBF-SEM. Two additional mice per group were euthanized at the same endpoint and conventional IHC was conducted. In the three remaining mice, the tumor was allowed to re-grow until it reached 100% of the tumor volume at the start of treatment, demonstrating these residual tumors had the capacity to regrow back to their original size.

Figure 1. PDX TNBC tumors persist following conventional chemotherapy treatments.

(a) The experimental approach used in the study. Images were adapted from BioRender. (b) SBF-SEM 3D rendering approach. (c) Tumor volumes were monitored biweekly following administration of chemotherapies to mice bearing orthotopic PIM001-P tumors (n = 6 mice/group). Arrows at the top indicate when chemotherapy was administered. Asterisks (*) above the x-axis indicate early euthanasia due to animal health concerns. Residual tumors were harvested, as noted by arrows. Error bars represent the standard error of the mean. (d) Residual tumors were harvested, processed for FFPE, and analyzed by H&E staining, (e) IHC for Ku80 and human-specific mitochondria. Scale bars are 50 μm.

H&E staining of FFPE tissue sections revealed poorly differentiated cellular architecture with abundant mitotic figures in the treatment-naïve tumors. All residual tumors in chemotherapy treatment groups showed extensive desmoplastic response with significant tumor cell pleomorphism, including atypical and enlarged nuclear and cytoplasmic sizes and shapes, pronounced nucleoli, and areas of necrosis, consistent with chemotherapy-induced cytological changes observed previously in PDXs and in patients^13,43 (Fig 1e). IHC staining with Ku80 to distinguish human tumor cells from mouse stromal cells revealed abundant stromal infiltration in all residual tumors (Fig. 1e), as shown previously with AC treatment¹³. IHC, using an antibody specific to human mitochondria (excluding mouse mitochondria), revealed increased staining in residual tumor cells compared to treatment-naïve cells, consistent with our prior findings for AC-treated residual tumors (Fig. 1e)¹³. Therefore, various conventional chemotherapy agents induced partial tumor regression in PIM001-P, and residual tumors persisting after these four treatments had aberrant histology architecture and increased mitochondrial content in the tumor cell compartment.

3D rendering of mitochondria by SBF-SEM was used to quantify mitochondrial features.

From each treatment group, we selected one representative tumor with a volume near each group’s mean tumor volume to generate blocks for SBF-SEM. Electron micrographs were taken at either 15 nm × 15 nm × 50 nm or 10 nm × 10 nm × 50 nm resolution with a minimum of three ROIs, each with 300 slices per sample. We inspected ROIs prior to imaging to avoid necrotic and stromal tissue. Mitochondria segmentation was conducted on different areas of the orthogonal (ortho) slice (Fig. 2a–e) to include any spatial heterogeneity within each tumor. Segmented mitochondria used for final analyses were each analyzed manually to ensure they were completely segmented (i.e., not missing a slice) (Fig. 2a’–e’). From each tumor, we randomly selected 350 mitochondria for further analysis (Fig. 2a’–e”). For each mitochondrion, we computed area, volume, perimeter, width, length, mitochondria complex index, and mitochondria branching index as described previously^17,44.

Figure 2. Several mitochondrial features were significantly altered by single-agent chemotherapy treatments.

(a-e) Representative SBF-SM orthoslices. (a’-e’) Representative 3D SBF-SEM orthoslices with overlays of mitochondria segmentations. (a”-e”) Representative 3D SBF-SEM mitochondria segmentations for each treatment group: treatment-naïve (TN), Adriamycin + cyclophosphamide (AC), docetaxel + carboplatin (DTX+CRB), docetaxel (DTX), and carboplatin (CRB). The scale bar is 3 μm. Mitochondrial measurements were calculated using Amira software for 350 segmented mitochondria (represented as dots) from each tumor for (f) volume, (g) 3D area, (h), and perimeter. Adjusted p-values (*p < 0.05, **p < 0.01, ***p < 0.001, ****p<0.0001) are shown on box and whisker plots as determined using the Mann–Whitney U test followed by the Holm method.

We compared the mean values of each mitochondrial feature between tumors. To test for differences in the mean of a measured value across treatment groups, we performed a Mann–Whitney U test because q–q plots revealed that mitochondrial features were non-normally distributed for all treatment groups, precluding the use of ANOVA (see Supplementary Fig. 1). The single-agent treatment groups, DTX and CRB, had significantly increased mitochondrial area, volume, and perimeter compared to treatment-naïve tumors (Fig 2f–h), which was not observed for residual tumors after either of the combination therapies. These trends were also apparent in density plots (see Supplementary Fig. 2 & 3), which further illustrate the distinct distribution of mitochondrial features observed between each tumor. Thus, residual tumors that persisted after single-agent chemotherapies had significantly increased mitochondrial length, area, volume, and perimeter compared to the treatment-naïve mitochondria. In contrast, most of these measurements did not increase significantly in residual tumors after treatment with combination therapies. Of note, several significant structural differences were observed between residual mitochondria following single-agent versus combination chemotherapy treatment.

Residual tumors exhibited increased mitochondrial complexity.

We then analyzed mitochondria for complex measurements, including sphericity and mitochondrial branching index (MBI). Transverse and longitudinal views of the segmented mitochondria were analyzed to determine mitochondrial complexity (Fig. 3a). Sphericity scores range from 0–1, with 1 representing a perfect sphere (Fig. 3c). MBI compares transverse (length) to longitudinal (width) measurements to determine the directionality of any branching (Fig. 3b). For example, if MBI is greater than 1, then the width measurement is greater than the length measurement, and the branching is along the x-axis. Conversely, if the MBI is less than 1, the branching is along the y-axis. If MBI is around 1, then the branching along the x-axis and y-axis is similar. Single-agent chemotherapy-treated mitochondria had significantly decreased sphericity and MBI compared to treatment-naïve tumors, consistent with increased narrow, branched mitochondria. The combination treatment groups had an increased MBI and sphericity compared to the treatment-naïve group, consistent with less branched mitochondria and more spherical mitochondria (Fig. 3b–c).

Figure 3. Single chemotherapeutic agent treated residual tumors have the greatest mitochondrial complexity.

Mitochondrial networks revealed through 3D mitochondria rendered using SBF-SEM. (a) Representative longitudinal and transverse views of the segmented mitochondria for each treatment group: treatment-naïve (TN), Adriamycin + cyclophosphamide (AC), docetaxel + carboplatin (DTX+CRB), docetaxel (DTX), and carboplatin (CRB). The scale bar is 3 μm. Mitochondrial measurements were calculated using Amira software for 350 segmented mitochondria (represented as dots) from each tumor for (b) mitochondrial branching index (MBI) and (c) sphericity, along with their respective method of measurements below. Adjusted p-values (*p < 0.05, **p < 0.01, ***p < 0.001, ****p<0.0001) are shown on box and whisker plots as determined using the Mann–Whitney U test followed by the Holm method.

The extensive intra-tumoral structural heterogeneity of treatment-naïve mitochondria was reduced in residual tumors.

ITH is a well-documented characteristic of TNBC and other cancers^35,45, yet the specific heterogeneity within mitochondrial structures remains unexplored in breast cancer. To address this, we quantified mitochondrial ITH by calculating the variance for each mitochondrial feature between all 350 mitochondria within each individual tumor, then compared these variances between tumors using an F-test (see Supplementary Fig. 4a). This revealed that the treatment-naïve tumor exhibited significantly higher variance in all mitochondrial features compared to residual tumors post-chemotherapy (see Supplementary Fig. 4b–i). Specifically, the variance in mitochondrial area, volume, perimeter, width, length, and MCI was significantly greater in treatment-naïve tumors (see Supplementary Fig. 4b–i). Furthermore, combination chemotherapy treatments resulted in a greater reduction in the variance of area, perimeter, width, and MCI compared to single-agent treatments (see Supplementary Fig. 4c&d, 4g&h). These findings indicate that there is significant 3D mitochondrial ITH in this PDX model of TNBC, and that chemotherapy, particularly combination regimens, drastically reduced this heterogeneity.

Mito-otyping revealed heterogeneous and complex mitochondrial structures in treatment-naïve and residual tumors.

To visualize the complexity of mitochondrial structures, we performed mito-otyping²⁴, or karo-type-like arrangement of mitochondria to visualize complexity relative to volume (Fig. 4a). As a result of the extensive ITH of mitochondria, we displayed a representative subset of the top, middle, and bottom 10% of mitochondria, based on volume, within each tumor. Consistent with the quantitative data described above, we found extensive ITH in each specimen with a shift in the size and complexity of mitochondrial structures after chemotherapy. We computed the mitochondrial complexity index (MCI), measuring mitochondria as a 3D object by treating surface area as a function of volume^17,24 (Fig. 4b). In agreement with the aforementioned mitochondrial alterations, single-agent chemotherapy-treated mitochondria had significantly greater MCI than treatment-naïve mitochondria (Fig. 4b). Combination-treated mitochondria showed a non-significant trend towards decreased MCI compared to treatment-naïve mitochondria.

Figure 4. Mito-otyping of the diverse mitochondria structures observed within each treatment group.

(a) Representative mitochondria structures from the bottom, middle, and top 10% of the 3D volume for tumors are displayed. Scale bars are 1μm. (b) Diagram of the mitochondrial measurements used to characterize mitochondria structure and mitochondria network through the mitochondrial complex index (MCI), as calculated by Amira software for 350 segmented mitochondria (represented as dots) from each tumor. The scale bar is 3 μm. Equations for these measurements are shown in Fig. 4b. Adjusted p-values (*p < 0.05, **p < 0.01, ***p < 0.001, ****p<0.0001) on the box and whisker plots were determined using the Mann–Whitney U test, followed by the Holm method.

DISCUSSION

Although chemotherapy is the preeminent treatment modality for neoadjuvant management of TNBC, molecular rationale for choice of specific chemotherapeutic regimens is lacking. Although there are efforts to identify underlying molecular features to inform targeted treatments, such as with PARP inhibitors², immune checkpoint inhibitors⁴⁶, and predictive biomarkers^42,47, there is still little information on the chemotherapies that would be most effective for a patient based on the molecular features of the tumor. Thus, it is critically important to elucidate the biological characteristics of tumor cells that survive NACT. Improving chemotherapeutic responses has the potential to improve long-term prognoses for those ~45% of TNBC patients whose tumors persist despite enduring months of harsh chemotherapeutic exposure.

We previously established the importance of mitochondria structure dynamics in supporting oxphos in chemoresistant TNBC. We further demonstrated that inhibition of oxphos¹³ or mitochondrial fusion¹¹ in residual tumors persisting after chemotherapy exposure was able to significantly prolong tumor control. Although we and others demonstrated that the cells that survive some conventional chemotherapies rely heavily on mitochondrial oxphos¹¹, little is known about mitochondrial complexity in TNBC, and 3D mitochondrial structures remain unexplored in this disease. Our study provides evidence that the mitochondrial network is substantially altered in residual tumors, and that this is associated with the type of chemotherapeutic regimen administered.

We observed greater area, perimeter, length, and MCI in the mitochondria of residual tumors after single-agent chemotherapy than in the mitochondria of treatment-naïve tumors. These mitochondria also had greater volume and width compared to mitochondria from tumors after combination chemotherapy. Thus, the mitochondria in residual tumors after single-agent chemotherapies may support increased metabolic capacity, as evidenced by our previous study of TNBC showing that mitochondrial elongation led to enhanced oxphos in residual tumors¹³. Mitochondrial dysfunction, a complex condition with a wide range of health consequences, cannot be definitively determined from the 3D mitochondrial structure. However, given the importance of understanding mitochondrial dynamics in TNBC¹¹, the unique mitochondrial shapes we observed in this study merit further functional investigations that may lead to therapeutic insights.

Mitochondria interact and communicate with the endoplasmic reticulum (ER), and mitochondrial fission occurs at mitochondria-ER contact sites. Mitochondria also interact with the cellular machinery, including lipid droplets, the nucleus, Golgi, and lysosomes^19,48–50. However, the prevalence and importance of these mitochondria-organelle contact sites are poorly understood in cancer and remain drastically under-explored in TNBC⁵¹. We observed increased surface area in the mitochondria of residual tumors after single-agent chemotherapy, which could increase opportunity for interactions with other organelles. Future structural studies mapping the connections between mitochondria and key organelles, such as the ER or lipid droplets, will provide critical insights into these important subcellular mechanisms of communication.

We found that mitochondrial structural adaptations were more pronounced in tumors treated with a singleagent compared to those with a combination treatment. Combinations of chemotherapies or chemotherapies with other drugs can improve treatment efficacy, but there is evidence for antagonism or weak additive effects^52–54. Therefore, it is crucial to understand mechanisms of resistance to individual therapies that make up the backbone of our standard cancer treatments. Differences in mitochondria in tumors treated with a single agent vs. a combination therapy may indicate a dose-dependency, i.e., mitochondria that adapt their shape in response to single agents cannot adapt when cells receive multiple agents simultaneously. This is important because multiple chemotherapies can be given to TNBC patients either simultaneously or sequentially as part of a standard neoadjuvant regimen. Our data, which suggest that the scheduling of chemotherapies affects the ability of mitochondria to adapt, will be important for the success of future mitochondria-targeting therapies developed for translation in TNBC patients.

We observe extensive ITH in mitochondrial structure, especially in the treatment-naïve tumors. Mitochondrial structural heterogeneity is tissue-specific and is determined by the need for mitochondrial activity; for example, calf muscles that are a mixture of mitochondria-rich and mitochondria-poor fibers have greater mitochondrial structural heterogeneity than cardiac muscle that is in constant motion and require continuous ATP²⁴. The mitochondria of neuronal axons and dendrites have different morphologies that match the different energy demands of these neuronal compartments²⁷. Thus, the ITH in the treatment-naïve tumor may reflect the spatial ITH of many proteins and other molecules that have been found in human TNBC specimens^55–57. Heterogeneous mitochondria may provide energetic adaptability for cell survival in the face of stressors such as chemotherapy, which will need to be determined experimentally. All chemotherapy treatments we tested greatly reduced mitochondrial ITH, suggesting that mitochondria may undergo a ‘selective bottleneck’ to improve cell survival after the assault of chemotherapy. Indeed, many studies in breast cancer have provided evidence for enrichment of a cancer stem cell-like, or ‘tumor-initiating cell,’ subpopulation after a variety of therapeutic exposures^58,59. Another potential mechanism underpinning these changes is dynamic adaptation of mitochondrial structure that arises within cells following chemotherapeutic exposure but is not necessarily restricted to a subpopulation or subclone. There is substantial evidence in breast cancer for this mechanism as well^8,13,60,61. Indeed, mitochondria are incredibly dynamic, with fusion and fission coordinated with multiple cellular events such as cell cycle. It will be important that future investigations specifically investigate mitochondrial structure and function in this context in TNBC.

We showed previously in several TNBC cell lines that DNA-damaging chemotherapies such as CRB and Adriamycin induce mitochondrial fusion and that genetic or pharmacological inhibition of OPA1 increases chemosensitivity and reduces oxphos¹¹. Conversely, taxane chemotherapies (including DTX) do not induce fusion or oxphos. Our present study revealed a greater increase in mitochondrial area, volume, perimeter, length, and MCI in mitochondria from CRB-treated tumors compared to DTX-treated tumors. It will be important to validate this finding in additional TNBC models, as this chemotherapy-specific effect on mitochondria could inform mitochondrial-targeting therapy development.

Mitochondria are not static organelles, with cycles of fission and fusion supporting their dynamic nature. Our imaging only captured a snapshot of mitochondrial structures in our tumor samples, and we suspect this may represent the ‘average’ mitochondrial state in each tumor specimen. As 3D methods with subcellular resolution evolve for longitudinal, or even real-time analyses, the potential may arise to evaluate rates of mitochondrial structure changes in experimental models. As we showed previously, both pharmacologic and genetic perturbations can induce these mitochondrial structures¹¹. It is reasonable to speculate that these mitochondrial structures are plastic and can reverse, as reported prior in this PDX model, PIM001-P, treated with AC that there is a reversible drug-tolerant state¹³. In this previous study, barcode-mediated clonal tracking was used and did not provide evidence for clonal enrichment¹³.

Our study has several limitations. While SBF-SEM microscopy provides detailed information on the 3D landscape of subcellular organelles, it is exceptionally labor- and time-intensive, requiring specialized expertise¹⁶, which precludes its routine use in research labs and the clinic. For this reason, we herein conducted an in-depth analysis of mitochondria for one tumor per treatment group, providing spatially distinct ROIs for mitochondrial quantifications to cover the possible spatial heterogeneity within each tumor. It remains to be determined how reproducible mitochondrial features will be across large cohorts of mouse or human tumors. However, based on the extensive similarities between the two single-agent groups as well as between the two combination groups, we suspect the hundreds of mitochondria sampled within each tumor are a fair representation of the true disease. Although there are efforts to automate the analyses of the mitochondria and other organelles, algorithms often need to be retrained on new data sets⁶². It is also critically important to expand these studies to additional TNBC specimens from other patients because inter-patient heterogeneity within TNBC is well documented at the DNA, RNA, and protein levels^56,63,64. There is interest in including 3D modeling to provide a more comprehensive understanding of cancer progression and drug responses^65,66. Mitochondrial structural differences among TNBC subtypes are not yet characterized but will be important in future research studies. Finally, PDX models lack a fully intact immune system, so it will be important to follow up these studies on bona fide human tumor specimens. Nonetheless, publications from ours and many other groups provide evidence that PDX models accurately capture many features of human cancer, including ITH, metabolic phenotypes, and chemotherapeutic responses^30,42,67.

Supplementary Material

Supplement 1

Supplementary Table 1. Selected segmented mitochondria used for analyses. Number of segmented mitochondria characterized during this study and the process for selecting 350 mitochondria per treatment for analyses.

Supplement 2

Supplementary Figure 1. Mitochondrial measurements across treatment groups do not have normal distributions. Q–q plots of the raw measurements across all tumors are shown. Each blue dot represents one of the segmented mitochondria from a total of 350 segmented mitochondria. The red line indicates a reference for where the normal distribution falls on the graph.

Supplement 3

Supplementary Figure 2. Distribution of mitochondrial metrics within each treatment group. Density plots are of log10 transformed data showing the distribution of the respective metric for the segmented mitochondria. The line indicates the median for treatment-naïve tumors. Treatment groups: treatment-naïve (TN), Adriamycin + cyclophosphamide (AC), docetaxel + carboplatin (DTX+CRB), docetaxel (DTX), and carboplatin (CRB)

Supplement 4

Supplementary Figure 3. Metrics width and length were used to calculate MBI for each treatment group. Mitochondrial measurements for width and length were calculated using Amira software for 350 segmented mitochondria (represented as dots) from each treatment group. (a) Width (b) Length. Adjusted p-values (*p < 0.05, **p < 0.01, ***p < 0.001, ****p<0.0001) are shown on box and whisker plots as determined using the Mann–Whitney U test followed by the Holm method. Density plots are of log10 transformed data showing the distribution of the respective metric for the segmented mitochondria. The line indicates the median for treatment-naïve tumors. (c) Density plot for width (d) Density plot for length. Treatment groups: treatment-naïve (TN), Adriamycin + cyclophosphamide (AC), docetaxel + carboplatin (DTX+CRB), docetaxel (DTX), and carboplatin (CRB).

Supplement 5

Supplementary Figure 4. Significant ITH was observed in treatment-naïve mitochondria, while ITH is reduced upon chemotherapy treatment. (a) Schematic for comparing the variance between treatment groups across all mitochondrial features. (b-i) Variance bar graphs for all 350 segmented mitochondrial volume, area, perimeter, width, length, mitochondrial branching index (MBI), sphericity, and mitochondrial complex index (MCI). F-test and Holm method for adjusted p-values (*p < 0.05, **p < 0.01, ***p < 0.001, ****p<0.0001).

Funding Statement

GVE is a CPRIT Scholar in Cancer Research. Funding sources that supported this work include the Cancer Prevention and Research Institute of Texas RR200009 (to GVE); NIH 1K22CA241113-01 (to GVE); NIH T32 predoctoral training grant T32GM136560-02 (to MJB); BWF Ad-hoc Award, National Institutes of Health (NIH) Small Research Pilot Subaward to 5R25HL106365-12 from the NIH PRIDE Program, DK020593 (to AHJ); 1R37CA269783-01A1 (to GVE); a BCM Breast Program SPORE Career Enhancement Program Grant (to GVE); National Science Foundation Graduate Research Fellowship 2140736 (to MJB); American Cancer Society Research Scholar Grant RSG-22-093-01-CCB (to GVE), a Myra Branum Wilson Baylor Research Advocates for Student Scientists Scholarship (to MJB); and a Breast Cancer Alliance Young Investigator Grant (to GVE); UNCF/Bristol-Myers Squibb (UNCF/BMS) E.E. Just Postgraduate Fellowship in Life Sciences Fellowship and Burroughs Wellcome Fund/ PDEP #1022376 (to HKB) UNCF/BMS E.E. Just Faculty Fund, Career Award at the Scientific Interface (CASI Award) from Burroughs Welcome Fund ID # 1021868.01, Vanderbilt Diabetes and Research Training Center for Department of Medicine’s Diabetes Research and Training Center (DRTC) Alzheimer’s Disease Pilot & Feasibility Program, CZI Science Diversity Leadership grant number 2022-253529 from the Chan Zuckerberg Initiative DAF, and an advised fund of Silicon Valley Community Foundation (to AHJ). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health, the Cancer Prevention and Research Institute of Texas, National Science Foundation, American Cancer Society, BRASS, UNCF/BMS, Burroughs Welcome Fund, DRTC, Chan Zuckerberg Initiative DAF, or Silicon Valley Community Foundation.

DATA AND CODE AVAILABILITY

Mitochondria images and raw data are available from the corresponding authors upon request. All computational codes used for statistical analyses are available https://github.com/audra-lane/berner-et-al-2024. The PDX model PIM001-P can be made available through a materials transfer agreement with The University of Texas MD Anderson Cancer Center.