Quantitative Optical Imaging of Primary Tumor Organoid Metabolism Predicts Drug Response in Breast Cancer

With the ever-increasing number of drugs approved to treat cancers, selection of the optimal treatment regimen for an individual patient is challenging. Physicians weigh the potential benefits of the drugs against the side-effects to the patient. Currently, drug regimens for breast cancer are chosen on the basis of tumor expression of several proteins, including estrogen receptor (ER), progesterone receptor, and high levels of human epidermal growth factor receptor 2 (HER2), assessed in the diagnostic biopsy, and drug effectiveness is determined after weeks of treatment from tumor size measurements. A personalized medicine approach would identify the optimal treatment regimen for an individual patient and reduce morbidity from overtreatment.

Current methods to assess therapy response include tumor size, measured by mammography, MRI, or ultrasound. These methods evaluate the regimen that the patient received. Molecular changes induced by antitumor drugs precede changes in tumor size and may provide proximal endpoints of drug response. Cellular metabolism may provide biomarkers of early treatment response, because oncogenic drivers typically affect metabolic signaling (1, 2). Indeed fluoro-deoxy-glucose (FDG)–PET has been explored as a predictor of response but lacks the resolution and sensitivity to accurately predict therapy response on a cellular level (3, 4).

Optical metabolic imaging (OMI) provides unique sensitivity to detect metabolic changes that occur with cellular transformation (5–10) and upon treatment with anticancer drugs (11). OMI uses the intrinsic fluorescence properties of NADH and FAD, coenzymes of metabolic reactions. OMI endpoints include the optical redox ratio (the fluorescence intensity of NADH divided by the fluorescence intensity of FAD), the NADH and FAD fluorescence lifetimes, and the “OMI index” (a linear combination of these three endpoints). The optical redox ratio provides a dynamic readout of cellular metabolism (12), with increased redox ratio (NADH/FAD; ref. 8) observed in malignant cells exhibiting the Warburg effect (increased glycolysis despite the presence of oxygen; ref. 13). Fluorescence lifetime values report differences in fluorophore conformation, binding, and microenvironment, such as pH, temperature, and proximity to quenchers such as free oxygen (14). OMI endpoints report early, molecular changes due to anticancer drug treatment (11), and are powerful biomarkers of drug response.

Primary tumors can be cultured ex vivo as organoids, which contain the malignant tumor cells and the supporting cells from the tumor environment, such as fibroblasts, leukocytes, endothelial cells, and hematopoietic cells (15). Interactions between cancer cells and stromal cells have been shown to mediate therapeutic resistance in tumors (16). Therefore, organoid cultures provide an attractive platform to test cancer cell response to drugs in a relevant, “body-like” environment. Furthermore, multiple organoids can be generated from one biopsy, enabling high-throughput tests of multiple drug combinations with a small amount of tissue.

OMI of primary tumor organoids enables high-throughput screening of potential drugs and drug combinations to identify the most effective treatment for an individual patient. Here, we validate OMI in primary tumor organoid cultures as an accurate, early predictor of in vivo tumor drug response in mouse xenografts, and present the feasibility of this approach on primary human tissues. The cellular resolution of this technique also allows for subpopulations of cells to be tracked over time with treatment, to identify therapies that affect all cells in a heterogeneous population.

This study was approved by the Vanderbilt University Animal Care and Use Committee and meets the NIH guidelines for animal welfare. BT474 cells or HR6 cells (10⁸) in 100 μL Matrigel were injected in the inguinal mammary fat pads of female athymic nude mice (J:NU; The Jackson Laboratory). Tumors grew to ≥200 mm³. Tumor-bearing mice were treated twice weekly with the following drugs: control human IgG (10 mg/kg, i.p.; R&D Systems), trastuzumab (10 mg/kg, i.p.; Vanderbilt Pharmacy), paclitaxel (2.5 mg/kg, i.p.; Vanderbilt Pharmacy), XL147 (10 mg/kg, oral gavage; Selleck Chemicals), trastuzumab + XL147, trastuzumab + paclitaxel, and trastuzumab + paclitaxel + XL147. Tumor volume was calculated from caliper measurements of tumor length (L) and width (W), (L × W²)/2, twice a week.

This study was approved by the Vanderbilt University Institutional Review Board and informed consent was obtained from all subjects. A primary tumor biopsy, removed from the tumor mass after surgical resection, was provided by an expert breast pathologist (M.E. Sanders). The tumor was placed immediately in sterile DMEM, transported on ice to the laboratory (∼5-minute walk), and generated into organoids within 3 hours of tissue resection. Pathology and receptor status of the tissue were obtained from the patient's medical chart.

Breast tumors (xenografts and primary) were washed three times with PBS. Tumors were mechanically dissociated into 100 to 300 μm macrosuspensions in 0.5 mL primary mammary epithelial cell (PMEC) media [DMEM:F12 + EGF (10 ng/mL) + hydrocortisone (5 μg/mL) + insulin (5 μg/mL) + 1% penicillin:streptomycin] by cutting the tissues with a scalpel or by spinning in a C-tube (Miltenyi Biotec). Macrosuspension solutions were combined with Matrigel in a 1:2 ratio, and 100 μL of the solution was placed on cover slips. The gels solidified at room temperature for 30 minutes and then for 1 hour in the incubator. The gels were over-lain with PMEC media supplemented with drugs. The following in vitro drug dosages were used to replicate in vivo doses (17–19): control (control human IgG + DMSO), trastuzumab (25 μg/mL), paclitaxel (0.5 μmol/L), XL147 (25 nmol/L), tamoxifen (2 μmol/L), fulvestrant (1 μmol/L), and A4 (10 μg/mL; Takis Biotech, Inc.).

Fluorescence lifetime imaging was performed on a custom-built multiphoton microscope (Bruker), as described previously (11, 20). Excitation and emission light were coupled through a 40× oil immersion objective (1.3 NA) within an inverted microscope (Nikon; TiE). A titanium:sapphire laser (Coherent Inc.) was tuned to 750 nm for NADH excitation (average power, 7.5–7.9 mW) and 890 nm for FAD excitation (average power, 8.4–8.6 mW). Bandpass filters, 440/80 nm for NADH and 550/100 nm for FAD, isolated emission light. A pixel dwell time of 4.8 microseconds was used to acquire 256 × 256 pixel images. Each fluorescence lifetime image was collected using time-correlated single-photon counting electronics (SPC-150; Becker and Hickl) and a GaAsP PMT (H7422P-40; Hamamatsu). Photon count rates were maintained above 5 × 10⁵ for the entire 60-second image acquisition time, ensuring that no photobleaching occurred. The instrument response full width at half maximum was 260 picoseconds as measured from the second harmonic generation of a urea crystal. Daily fluorescence lifetime validation was confirmed by imaging of a fluorescent bead (Polysciences Inc). The measured lifetime of the bead (2.1 ± 0.06 nanoseconds) concurs with published values (10, 20, 21).

Fluorescence lifetime images of organoids were acquired at 24, 48, and 72 hours after drug treatment. Organoids were grown in 35-mm glass-bottom Petri dishes (MatTek Corp) and imaged directly through the coverslip on the bottom of the Petri dish. Six representative organoids from each treatment group were imaged. The six organoids imaged contained collectively approximately 60 to 300 cells per treatment group for statistical and subpopulation analyses. First, an NADH image was acquired and a subsequent FAD image was acquired of the exact same field of view.

A previously reported protocol (22) was adapted for immunofluorescent staining of organoids. Briefly, gels were washed with PBS and fixed with 2 mL 4% paraformaldehyde in PBS. Gels were washed with PBS, and then 0.15 mol/L glycine in PBS was added for 10 minutes. Gels were washed in PBS, and then added to 0.02% Triton X-100 in PBS. Gels were washed with PBS then overlain with 1% fatty acid–free BSA, 1% donkey serum in PBS. The next day, the solution was removed and 100 μL of antibody solution (diluted antibody in PBS with 1% donkey serum) was added to each gel. The gels were incubated for 30 minutes at room temperature, washed in PBS three times, and then incubated in 100 μL of secondary antibody solution for 30 minutes at room temperature. The gels were washed in PBS three times, washed in water twice, and then mounted on slides using 30 μL of the ProLong Antifade Solution (Molecular Probes).

The primary antibodies used were anti-cleaved caspase-3 (Life Technologies) and anti-Ki67 (Life Technologies). Both were diluted at 1:100. A goat anti-rabbit IgG FITC secondary antibody was used (Life Technologies). FITC fluorescence was obtained by excitation at 980 nm on the multiphoton microscope described above, and a minimum of six organoids were imaged. Positive staining of cleaved caspase-3 and Ki67 was confirmed by staining mouse thymus and mouse small intestine, respectively. Immunofluorescence images were quantified by manual counting of the total number of cells and the number of positively stained cells in each field of view. Immunofluorescence results were presented as percentage of positively stained cells, quantified from six organoids, approximately 200 cells.

Photon counts for nine surrounding pixels were binned (SPCImage). Fluorescence lifetime components were extracted from the photon decay curves by deconvolving the measured system response and fitting the decay to a two-component model, |$I(t) = \alpha _1 \exp ^{ - t/\tau _1} + \,\alpha _2 \exp ^{ - t/\tau _2 } + \,C$|⁠, where I(t) is the fluorescence intensity at time t after the laser pulse, α₁ and α₂ are the fractional contributions of the short and long lifetime components, (i.e., α₁ +α₂ = 1), τ₁ and τ₂ are the fluorescence lifetimes of the short and long lifetime components, and C accounts for background light. A two-component decay was used to represent the lifetimes of the free and bound configurations of NADH and FAD (10, 23, 24) and yielded the lowest χ² values (0.99–1.1), indicative of an optimal fit. Matrices of the lifetime components were exported as ascii files for further processing in Matlab.

To streamline the cellular-level processing of organoid images, an automated image analysis routine, as previously described (25), was used in Cell Profiler in Matlab. Briefly, a customized threshold code identified pixels belonging to nuclear regions that were brighter than background but not as bright as cell cytoplasms. These nuclear pixels were smoothed and the resulting round objects between 6 and 25 pixels in diameter were segmented and saved as the nuclei within the image. Cells were identified by propagating out from the nuclei. An Otsu Global threshold was used to improve propagation and prevent propagation into background pixels. Cell cytoplasms were defined as the cells minus the nuclei. Cytoplasm values were measured from each OMI image (redox ratio, NADH τ_m, NADH τ₁, NADH τ₂, NADH α₁, FAD τ_m, FAD τ₁, FAD τ₂, and FAD α₁).

The redox ratio, NADH τ_m, and FAD τ_m were norm-centered across cell values from all treatment groups within a sample, resulting in unitless parameters with a mean of 1. The OMI index is the linear combination of the norm-centered redox ratio, NADH τ_m, and FAD τ_m, with the coefficients (1, 1, and −1), respectively, computed for each cell. The three endpoints, redox ratio, NADH τ_m, and FAD τ_m are independent variables (11) and are thus weighted equally. The signs of the coefficients were chosen to maximize difference between control and drug-responding cells.

Subpopulation analysis was performed by generating histograms of all cell values within a group as previously reported (11). Each histogram was fit to 1, 2, and 3 component Gaussian curves. The lowest Akaike information criterion (AIC) signified the best fitting probability density function for the histogram (26). Probability density functions were normalized to have an area under the curve equal to 1.

Differences in OMI endpoints between treatment groups were tested using a student t test with a Bonferroni correction. An α significance level less than 0.05 was used for all statistical tests.

Validation of an organoid-OMI screen for drug response was first tested in two isogenic HER2-amplified breast cancer xenografts. BT474 xenografts are sensitive to the HER2 antibody trastuzumab, whereas HR6 xenografts, derived as a subline of BT474, are trastuzumab resistant. The following single drugs and drug combinations were tested: paclitaxel (P; chemotherapy), trastuzumab (H; anti-HER2 antibody), XL147 (X; PI3K small-molecule inhibitor; ref. 27), H+P, H+X, and H+P+X. Paclitaxel and trastuzumab are standard-of-care drugs, and XL147 is in clinical trials and preclinical studies support combination therapy of XL147 with trastuzumab for patients who have developed a resistance to trastuzumab (27, 28).

Representative redox ratio, NADH τ_m, and FAD τ_m images of BT474 xenograft-derived organoids demonstrate mixed multicellular morphology and highlight the subcellular resolution of this technique (Fig. 1A–F). A longitudinal study of tumor growth demonstrated that the BT474 xenografts responded to each treatment arm (Fig. 1G), with significant reduction in tumor volume, as determined from caliper measurements, on day 7 for all treatment groups except trastuzumab, which had significant reduction on day 11 (Fig. 1H).

A composite endpoint, the OMI index, was computed as a linear combination of the mean-normalized optical redox ratio, NADH τ_m, and FAD τ_m for each cell. After 24 hours of treatment, the OMI index was significantly reduced in all treated BT474 organoids, compared with the control (P < 0.05; Fig. 1I). By 72 hours, the OMI index decreased further in all the treatment groups (P < 5 × 10⁻⁷, Fig. 1J). The redox ratio, NADH τ_m, and FAD τ_m values showed similar trends (Supplementary Fig. S1). Changes in short and long lifetime values and in the portion of free NADH or FAD contributed to the changes in τ_m (Supplementary Table S1).

The high-resolution capabilities of OMI allowed single-cell analysis and population modeling for quantification of cellular subpopulations with varying OMI indices. Visual inspection of cell morphology suggested that the majority of cells are tumor epithelial cells; stromal cells with obvious morphologic differences were eliminated from the analysis. Population density modeling of cellular distributions of the OMI index revealed two populations with high and low OMI index values in all of the BT474-treated organoids at 24 hours (Fig. 1K and Supplementary Fig. S2). By 72 hours, the XL147-, H+P-, H+X-, and H+P+X-treated organoids have a single population with narrower peaks (Fig. 1L and Supplementary Fig. S2). The trastuzumab-treated organoids have two populations at 72 hours, both lower than the mean OMI index of the control organoids (Fig. 1L). Immunofluorescent staining of cleaved caspase-3 and Ki67 of BT474 organoids treated for 72 hours confirmed increased apoptosis and decreased proliferation in drug-treated organoids, with the greatest increases in cell death with combined treatments (Fig. 1M and N).

Next, the OMI-organoid screen was tested on trastuzumab-resistant HR6 xenografts (29). Representative images show HR6 organoid morphology and spatial distributions of OMI endpoints (Fig. 2A–F). These HER2-overexpressing tumors had continued growth with trastuzumab treatment (Fig. 2G). Treatment with paclitaxel and XL147 initially caused HR6 tumor regression (P < 0.05 on day 10 for XL147 and on day 14 for paclitaxel) but then resumed growth (Fig. 2G and H). Mice treated with the H+P, H+X, and H+P+X combination therapies exhibited sustained HR6 tumor reduction (Fig. 2G and H).

After 24 hours of treatment, significant reductions in the OMI index were detected in HR6 organoids treated with paclitaxel, XL147, H+P, H+X, and H+P+X (P < 0.05; Fig. 2I). At 72 hours, the OMI index of the paclitaxel- and XL147-treated organoids was significantly greater than that of the control organoids (P < 0.05; ref. Fig. 2J), consistent with the recovery of HR6 tumor growth after prolonged therapy (Fig. 2G). The organoids treated with drug combinations (H+P, H+X, and H+P+X) continued to have significantly lower OMI index values (P < 10⁻⁶) at 72 hours, compared with untreated controls. Individual OMI endpoints showed similar trends (Supplementary Fig. S3 and Supplementary Table S2). Subpopulation analysis revealed two subpopulations in the OMI index for all treated groups except for trastuzumab at 24 hours (Fig. 2K and Supplementary Fig. S4). By 72 hours, the paclitaxel- and XL147-treated organoids had a single population (Fig. 2L and Supplementary Fig. S4). Immunofluorescent staining of cleaved caspase-3 of organoids treated for 72 hours revealed increased cell death in HR6 organoids treated with H+P, H+X, and H+P+X (P < 0.05, Fig. 2M). The percentage of Ki67-positive cells at 72 hours decreased with paclitaxel, H+P, H+X, and H+P+X treatment (P < 0.005, Fig. 2N).

We tested these methods on primary breast cancer biopsies obtained from surgical resection. Tumors were obtained fresh from deidentified mastectomy specimens not required for further diagnostic purposes, and dissociated into organoids within 1 to 3 hours postresection. Cancer drugs were added and organoids were imaged with OMI. Representative redox ratio, NADH τ_m, and FAD τ_m images (Fig. 3) demonstrate the varying morphology of organoids derived from ER⁺, HER2⁺, and triple-negative breast cancers (TNBC).

When quantified, the OMI endpoints differed between cancer subtypes. In immortalized cell lines, the redox ratio was elevated in ER⁺/HER2⁻ cells and was greatest in HER2⁺ cells (P < 5 × 10⁻⁵, Fig. 4A). Similarly, NADH τ_m was increased in immortalized ER⁺/HER2⁻ and HER2⁺ breast cancer cells as compared with TNBC cells (P < 5 × 10⁻⁸, Fig. 4B). FAD τ_m was greatest in ER⁺/HER2⁻ cells (P < 0.05, Fig. 4C). Overall, the OMI index was lowest in TNBC and greatest in HER2⁺ cells (P < 5 × 10⁻⁸, Fig. 4D), suggesting that HER2 and ER expression influence cellular metabolism.

Similar trends were observed for the OMI endpoints in organoids derived from primary breast tumor specimens cultured under basal conditions. The redox ratio was increased in organoids from ER⁺/HER2⁻ tumors and was greatest in HER2⁺/ER⁻ organoids (P < 5 × 10⁻¹²; Fig. 4E; Supplementary Table S3). Likewise, NADH τ_m increased with ER and HER2 expression (P < 5 × 10⁻⁸; Fig. 4F). FAD τ_m was increased in ER⁺ organoids and reduced in HER2⁺ organoids (P < 0.05, Fig. 4G). The OMI index was lowest for TNBC, and greatest for HER2⁺ organoids (P < 5 × 10⁻³; Fig. 4H).

Organoids were generated from four ER+ (HER2⁻) primary human tumors and treated with the chemotherapeutic drug paclitaxel, the selective ER modulator tamoxifen, the HER2 antibody trastuzumab, and the pan-PI3K inhibitor XL147. Organoids derived from the first ER⁺ tumor had significantly reduced OMI index values upon treatment with paclitaxel, tamoxifen, XL147, H+X, H+P+T, H+P+X, and H+P+T+X for 72 hours (P < 5 × 10⁻⁵, Fig. 5A). Immunofluorescence of cleaved caspase-3 showed increased cell death in parallel organoids treated for 72 hours with paclitaxel, tamoxifen, XL147, H+X, H+P+T, H+P+X, and H+P+T+X (Fig. 5B). Subpopulation analysis revealed less variability (narrower histogram peaks) within responsive treatment groups compared with the cells of control and trastuzumab-treated organoids (Fig. 5C and Supplementary Fig. S5). Corresponding OMI endpoints showed similar trends (Supplementary Fig. S6 and Supplementary Table S4).

Organoids derived from a second ER⁺ tumor responded similarly. The OMI index decreased upon treatment with paclitaxel, tamoxifen, H+P, P+T, and H+P+T at 72 hours (P < 5 × 10⁻⁵, Fig. 5D). Subpopulation analysis revealed a single population of control cells that shifted to lower OMI indexes with paclitaxel, tamoxifen, H+P, P+T, and H+P+T treatments (Fig. 5E and Supplementary Fig. S7). Corresponding OMI endpoints showed similar trends (Supplementary Fig. S8 and Supplementary Table S5).

The third and fourth ER⁺ clinical samples yielded organoids with variable responses to treatment. Organoids derived from the third patient had significant reductions in OMI index after 24 hours of treatment with tamoxifen, XL147, H+X, and H+P+T+X (P < 0.005, Fig. 5F). Subpopulation analysis revealed two populations with high and low OMI index values for the H+P- and paclitaxel-treated organoids (Fig. 5G and Supplementary Fig. S9). Two populations, both with mean OMI index values less than that of the control organoids, were apparent in the organoids treated with XL147 and with H+P+T+X (Fig. 5G and Supplementary Fig. S9). Organoids from the fourth ER⁺ patient had reduced OMI indices following treatment with XL147, H+P, H+X, H+P+X, H+P+T, and H+P+T+X for 72 hours (P < 0.01, Fig. 5H). Subpopulation analysis of cells from these organoids revealed single populations with shifted mean OMI indices for all treatments except tamoxifen, H+P, and H+X, which had two populations (Fig. 5I and Supplementary Fig. S10). Corresponding OMI endpoints showed similar trends (Supplementary Figs. S11–S12 and Supplementary Tables S6–S7).

OMI was also performed on organoids derived from HER2⁺ (ER⁻) and TNBC specimens. Organoids derived from the HER2⁺ primary tumor were treated with the ER downregulator fulvestrant, the HER2 antibody trastuzumab, and the anti-ErbB3 antibody A4 (30). The OMI index was significantly decreased in the organoids treated for 24 hours with trastuzumab and A4 (P < 0.005, Fig. 6A). Subpopulation analysis revealed shifts in the mean OMI index values with these treatments within a single population of cells (Fig. 6B). Organoids derived from the TNBC specimen were treated with tamoxifen, the HER2 antibody trastuzumab, and the combination of trastuzumab plus tamoxifen (H+T). No significant changes were observed with these treatments in TNBC organoids after 24 hours (P > 0.3, Fig. 6C). Subpopulation analysis revealed a single population of cells from TNBC organoids (Fig. 6D). Corresponding OMI endpoints showed similar trends (Supplementary Figs. S7 and S8 and Supplementary Tables S8 and S9).

Primary tumor organoids are an attractive platform for drug screening because they are grown from intact biopsies, thus maintaining the tumor cells within the same tumor microenvironment (15). OMI is sensitive to early metabolic changes, achieves high resolution to allow analysis of tumor cell heterogeneity, and uses endogenous contrast in living cells for repeated measurements and longitudinal studies (11). The OMI index is a holistic reporter of cellular metabolism because the redox ratio and NADH and FAD lifetimes are independent measurements (11). The mean lifetime captures not only changes in free-to-bound protein ratios but also preferred protein binding and relative concentrations of NADH to NADPH (31). Cancer drugs have been shown to downregulate certain metabolism enzymes; for example, trastuzumab downregulates lactate dehydrogenase in breast cancer, and paclitaxel-resistant cells have been shown to have more lactate dehydrogenase expression and activity (32). The OMI index captures these drug-induced changes in metabolism enzyme activity. Organoids remain viable with stable OMI endpoints in controlled culture conditions (33), thus making them an attractive system to evaluate tumor response to drugs. We used OMI to assess the response of primary breast tumor organoids to a panel of clinically relevant anticancer agents used singly or in combination. Early OMI-measured response in organoids (24–72 hours after treatment) corroborated with standard tumor growth curves in xenografts, and the feasibility of this approach was confirmed in organoids derived from primary human breast tumors.

The OMI index was first evaluated as a reporter of tumor response in organoids derived from BT474 (ER⁺/HER2⁺) xenografts. Significant reductions in OMI index upon treatment with paclitaxel, trastuzumab, XL147, and combinations thereof, at both 24 and 72 hours, correlated with reduction of tumor growth (Fig. 1). Biochemically, cellular rates of glycolysis, and NADH and FAD protein-binding decrease with drug treatment in responsive cells (32), resulting in decreased redox ratios and NADH τ_m, and increased FAD τ_m in agreement with the decreased OMI index observed in drug-treated BT474 organoids. Significant reductions in tumor growth occurred 7 to 11 days after treatment initiation, whereas the OMI index detected response 24 to 72 hours after treatment. Cellular analysis revealed an initial heterogeneous response among cells within organoids treated with paclitaxel and H+P at 24 hours, which, by 72 hours, became a uniform response. The heterogeneity of trastuzumab-treated BT474 organoids persisted over 72 hours, suggesting an intrinsic subpopulation more susceptible to acquire drug resistance. This heterogeneity was not seen in the combination treatments, suggesting the combination treatments trump this drug resistance–prone subpopulation. OMI-measured response corroborated with increased cell death and decreased proliferation due to single and combination drug–treated organoids, measured with destructive postmortem techniques. The XL147-treated BT474 organoids have a much lower OMI index at 72 hours, but only a modest increase in cleaved caspase-3 activity. The same decrease was not observed in the HR6 cells that have alternative metabolism pathways activated because of their acquired resistance to trastuzumab. The OMI index detects changes in cellular metabolism that predict drug efficacy, but do not necessarily correlate with IHC.

In the current standard of care, patients with innate drug resistance are not identified a priori. We tested the capabilities of OMI to predict drug resistance using trastuzumab-resistant HR6 (ER⁺/HER2⁺) tumors (29). XL147 is a novel PI3K inhibitor under investigation for combined therapy with trastuzumab to improve response of resistant tumors (27). Significant reductions in the OMI index of HR6 organoids treated for 72 hours identified drug combinations (H+X, H+P, and H+P+X) that induced a sustained reduction in tumor growth in vivo (Fig. 2). The reduction in tumor growth upon treatment with H+X was consistent with previous reports of greater antitumor effects of the combination over trastuzumab and XL147 alone (27). Subpopulation analysis revealed multiple responses within the HR6 organoids after treatment with single drugs and combinations, suggesting increased heterogeneity compared with the parental BT474 organoids.

The OMI index of paclitaxel- and XL147-treated HR6 organoids initially decreased at 24 hours, and then increased at 72 hours, mirroring the tumor growth in mice after prolonged therapy, and indicating that the adaptations that allow HR6 cells to survive trastuzumab treatment also affect response to additional drugs. This relapse of HR6 tumors treated with paclitaxel and XL147 was not apparent until 2 to 3 weeks of drug treatment; yet, the OMI index identified a resistant population within both paclitaxel- and XL147-treated organoids at 24 hours and showed a selection of this population by 72 hours. Subpopulation analysis of the paclitaxel- and XL147-treated HR6 organoids revealed heterogeneous responses at 24 hours, suggesting that OMI is capable of early detection of resistant cells within a heterogeneous tumor. These results indicate that OMI of primary tumor organoids is able to identify heterogeneous responses within tumors on a cellular level, and potentially guide therapy selection early for maximal response. The ability to detect innate resistance at a cellular level before treatment may provide leads for identification of drugs that target such refractory subpopulations before they are selected by the primary therapy.

We next examined the feasibility of this approach using fresh tumor biopsies obtained from primary tumor surgical resections. OMI measurements in vivo and corresponding measurements from freshly excised tissues within 8 hours of surgery are statistically identical (20), providing ample time for specimen acquisition and transport to the laboratory. The morphology of organoids differed among patients and within breast cancer subtypes (Fig. 3), demonstrating a greater heterogeneity within primary tumors compared with xenografts.

Previously published studies report differences in OMI endpoints due to the presence or absence of ER and HER2 (8, 11, 34). Both ER and HER2 signaling pathways can influence metabolism: ER by inducing increased glucose transport (1), and HER2 through activation of PI3K (2), among other signal transducers. We compared OMI endpoints from immortalized cells and human tissue-derived organoids of three subtypes of breast cancer: ER⁺, HER2-overexpressing, and TNBC. The OMI index of immortalized cell lines increased with ER expression and was highest in HER2 overexpressing cells (consistent with prior studies (11)), and these trends were replicated in organoids derived from primary human tumors. Notably, NADH τ_m was significantly increased (P < 0.05) in the HER2⁺ organoids compared with ER⁺ organoids, but this trend was not observed in the immortalized cell lines. This difference could be due to molecular changes induced by the immortalization process, media components, primary tumor heterogeneity, and/or the heterogeneity within a primary breast tumor. Regardless, the results shown (Fig. 4) suggest breast cancer subtypes, ER⁺, HER2⁺ and TNBC, have different OMI profiles.

Organoids derived from human breast tumors were treated with a panel of breast cancer drugs (Figs. 5 and 6). Differences in the drug response of these organoids suggest heterogeneity across ER⁺/HER⁻ tumors. Organoids from one of the four ER⁺ tumors did not exhibit reduced OMI indices after treatment with tamoxifen. Organoids derived from two of the four ER⁺ samples did not have reduced OMI indices after paclitaxel treatment. These variable responses are consistent with variable responses seen with these drugs in the clinic (35–38). None of the organoids had reduced OMI indices with trastuzumab, which is expected because the organoids were derived from HER2⁻ tumors. Generally, the OMI index was reduced further upon treatment with drug combinations, supporting the use of drug combinations clinically.

Subpopulation analysis revealed cells within organoid treatment groups that exhibit different OMI indices after treatment, suggesting that subpopulations of cells with different drug sensitivities preexist and develop within primary tumors. Some of these cells may represent the cancer stem-like population with increased renewal capacity, metastatic potential, and drug resistance (39). The populations of organoids derived from human tumors have more variability (broader population curves) than those derived from xenografts, reflecting an inherent greater heterogeneity within primary tumors. This corroborates previous reports (40) of greater intratumoral heterogeneity in primary tumors than in xenografts derived from clonal cell lines. Thus, OMI imaging allows identification of heterogeneous cellular response to drug treatment in a dynamic population, which potentially enables drug selection to maximize therapeutic efficacy.

Organoids derived from HER2⁺/ER⁻ and TNBC primary tumors have OMI responses consistent with their clinical characteristics: reduced OMI index with trastuzumab treatment and no change with fulvestrant (ER antagonist) treatment in the HER2⁺/ER⁻ organoids (41), and no OMI index reductions after treatment with trastuzumab or tamoxifen in the TNBC organoids (Fig. 6a and c; refs. 42, 43). HER3 is an emerging target for breast cancer (30, 44) and the anti-HER3 antibody A4 reduced the OMI index of HER2⁺/ER⁻ organoids.

The results of this study support the validity of OMI for monitoring organoid response to anticancer drugs. We demonstrate high selectivity of the OMI index to directly measure drug response of organoids derived from breast cancer xenografts to single anticancer drugs and their combinations, and validated OMI measured response with gold standard tumor growth in two xenograft models. We have shown that the OMI index measured in primary tumor organoids resolves response and nonresponse within 72 hours, compared with the 3 weeks required to resolve this response with tumor size measurements. Furthermore, we extend this approach and generate drug response information from organoids derived from three subtypes of primary human tumors, TNBC, ER⁺, and HER2⁺. The high resolution of OMI allows subpopulation analysis for identification of heterogeneous tumor response to drugs in dynamic tumor cell populations. Altogether, these results suggest that OMI of primary tumor organoids may be a powerful test to predict the action of anticancer drugs and tailor treatment decisions accordingly.

L. Aurisicchio has ownership interest (including patents) in Takis Biotech. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.J. Walsh, M.E. Sanders, M.C. Skala

Development of methodology: A.J. Walsh, M.C. Skala

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.J. Walsh, R.S. Cook, M.E. Sanders, C.L. Arteaga, M.C. Skala

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.J. Walsh, R.S. Cook, M.E. Sanders, M.C. Skala

Writing, review, and/or revision of the manuscript: A.J. Walsh, R.S. Cook, M.E. Sanders, L. Aurisicchio, G. Ciliberto, C.L. Arteaga, M.C. Skala

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.L. Arteaga

Study supervision: M.C. Skala

Other (providing some key reagents): L. Aurisicchio

Other (providing monoclonal antibodies used for a subset of experiments reported in the article): G. Ciliberto

The authors thank C. Nixon, W. Sit, M. Madonna, and B. Stanley for assistance.

A.J. Walsh was supported by a grant from the National Science Foundation (DGE-0909667). M.C. Skala received grants from the DOD BCRP (DOD-BC121998), the NIH/NCI (NIH R00-CA142888, NCI Breast Cancer SPORE P50-CA098131), and Vanderbilt. G. Ciliberto received a grant from Associazione Italiana per la Ricerca sul Cancro (AIRC-IG10334).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.