Heterogeneity in tumor chromatin-doxorubicin binding revealed by in vivo fluorescence lifetime imaging confocal endomicroscopy

We present an approach to quantify drug–target engagement using in vivo fluorescence endomicroscopy, validated with in vitro measurements. Doxorubicin binding to chromatin changes the fluorescence lifetime of histone-GFP fusions that we measure in vivo at single-cell resolution using a confocal laparo/endomicroscope. We measure both intra- and inter-tumor heterogeneity in doxorubicin chromatin engagement in a model of peritoneal metastasis of ovarian cancer, revealing striking variation in the efficacy of doxorubicin–chromatin binding depending on intra-peritoneal or intravenous delivery. Further, we observe significant variations in doxorubicin–chromatin binding between different metastases in the same mouse and between different regions of the same metastasis. The quantitative nature of fluorescence lifetime imaging enables direct comparison of drug–target engagement for different drug delivery routes and between in vitro and in vivo experiments. This uncovers different rates of cell killing for the same level of doxorubicin binding in vitro and in vivo.

In vitro study to quantify the contribution of doxorubicin emission in the confocal endomicroscope detection channel and its impact on the measured EGFP fluorescence lifetime The confocal endomicroscope (CEM) employs 488 nm excitation and a 520-550 nm detection band to detect GFP fluorescence. However, doxorubicin (DOX) is also excited weakly at 488 nm and its emission spectrum bleeds through into the CEM detection window. Supplementary Figure 1 shows the normalized emission spectra measured using the LSM of H1-EGFP with 488 nm excitation and DOX with 488 nm excitation together with the transmission profiles of the emission filters used for multiphoton LSM and CEM FLIM measurements. To quantify the relative contributions of GFP and doxorubicin to the fluorescence lifetimes measured with the CEM, we undertook an in vitro study using IGROV-1 cells expressing H1-EGFP.
First, we used the LSM system with two-photon excitation at 900 nm and the 465-495 nm band-pass emission filter to measure the change in H1-EGFP fluorescence lifetime upon treatment with doxorubicin due only to FRET upon doxorubicin binding to chromatin. Globally fitting a double exponential model to this FLIM dataset yielded a χ 2 goodness-of-fit parameter of less than 1.1 with 2330 ps (72%) and 770 ps (28%) for the H1-EGFP non-FRETing and FRETing lifetime components respectively.
To understand how the DOX fluorescence could contribute to fluorescence decay profiles measured in the CEM detection channel, we used the CEM system with 488 nm excitation and detection over 520-550 nm to measure the fluorescence decay profile of DOX within wild type IGROV-1 cell nuclei in vitro. We globally fitted to a double exponential decay model and obtained lifetime components of 1590 ps (35%) and 390 ps (65%) respectively with a χ 2 of less than 1.2.
In order to quantify the relative contributions of H1-EGFP and DOX measured with the CEM, we measured the emission spectrum (490-630 nm) of IGROV-1 cells expressing H1-GFP after treatment for 3 hours with 9 μM doxorubicin. This measurement was performed using 488 nm excitation with the emission being analyzed using the diffraction grating-based spectral imaging capability of a Zeiss LSM780 microscope and is represented by the blue curve of Supplementary Figure 2b. We also measured the emission spectra of IGROV-1 cells expressing H1-EGFP with no DOX treatment (yellow curve) and of wild-type IGROV-1 cells treated with 9 µM DOX for 3 hours (pink curve).
The combined fluorescence signal expected in the CEM detection band (520-550 nm), SCEM, depends on the concentrations and the relative brightness of the two fluorophores, i.e. Assuming the fluorescence signal from doxorubicin scales linearly with doxorubicin concentration, the expected signal SDOX due to doxorubicin can be estimated for an arbitrary doxorubicin concentration by the equation The CEM fluorescence decay that we expect to measure can be expressed as a linear sum of the H1-EGFP and DOX fluorescence decays detected in the 520-550 nm emission window. The total fluorescence signal as a function of time after excitation for an infinitely narrow pulse of excitation light can therefore be described by a multi-exponential decay of four components: 2 for H1-GFP and 2 for doxorubicin Here 1 and 2 represent the long and short lifetime components of H1-EGFP; 1 and 2 represent the long and short lifetime components of DOX; 1 represents the long lifetime fraction of GFP and 1 represents the long lifetime fraction of DOX. The quantity EGFP IGROV-1 represents the total fluorescence signal from H1-GFP for the expression level achieved in our stable cell line.
The values of 1 were determined to be 2330 ps for multiphoton measurements and 2460 ps for CEM measurements. The value used for 2 was taken to be 770 ps, as determined with the multiphoton LSM in vitro measurements of IGROV-1 cells expressing GFP-H1 using a 465-495 nm emission filter to exclude the doxorubicin signal. It was not possible to measure a DOX-cross-talk-free value of 2 with the CEM. As discussed above, 1 and 2 were determined to be 1590 ps and 390 ps respectively from CEM in vitro measurements.
Using this approach to determine the relative fractions of the H1-EGFP non-FRETing and To determine the intracellular DOX concentration from in vivo CEM measurements that may be compromised by the DOX bleed through, we compared the dose response curves for the in vitro measurements obtained with the multiphoton LSM and CEM instruments. The multiphoton LSM data was fitted to a double exponential decay model with 1 and 2 fixed to 2330 ps and 770 ps respectively and the CEM data was fitted to a double exponential decay model with 1 and 2 fixed to 2460 ps and 770 ps respectively. As shown in Supplementary Figure 3c that plots the % population fraction of the non-FRETing components, the curves are similar up to 1 µM DOX. Above this concentration, the CEM appears to underestimate the non-FRETing H1-EGFP contribution because of the DOX bleed through.
We used the in vitro data presented in Supplementary Figure 3c as a calibration to correct the in vivo CEM measurements. The in vivo CEM FLIM data was first fitted to a double exponential decay model with the non-FRETing and FRET component lifetimes fixed to 2376 ps and 770 ps respectively -noting that the non-FRETing H1-EGFP lifetime is that measured in vivo for control mice without doxorubicin treatment. The βG1 value obtained from this in vivo data can then be corrected according to a look-up table obtained by interpolating the data plotted in Supplementary Figure 3c. This corrected value of βG1 can then be used to estimate the intracellular DOX concentration that would have been measured in the absence of bleed through of the DOX fluorescence in the CEM spectral detection window.
To illustrate that the CEM measured DOX response data is not dominated by the DOX bleed through, fluorescence decay profiles were simulated for the 520-550 nm CEM detection band using the expressions above for doxorubicin concentrations in the range of 0 -20 µM, as used in vitro and expected in vivo.
Using the lifetime components for the in vitro CEM measurements (i.e. 2460 ps as measured with the CEM for non-FRETing H1-EGFP and 770 ps for FRETing H1-EGFP as measured without doxorubicin bleed through with the multiphoton LSM), Supplementary Figure 3d shows the simulated % of non-FRETing H1-EGFP in the detected CEM signal expected for the situation where there is no quenching of the H1-EGFP fluorescence but a contribution from DOX bleed-through (red curve) and where there is a contribution from quenched H1-EGFP fluorescence as well as the DOX contribution (blue curve). Also shown is the measured CEM signal as a function of DOX concentration (magenta curve). These results indicate that directly excited doxorubicin bleed through provides only a minor contribution to the CEM fluorescence lifetime measurements and the observed increase in the population fraction of FRETing H1-EGFP is the major factor in the reduction of fluorescence lifetime response to doxorubicin measured with the CEM.

Potential sources of error in the fluorescence lifetime analysis of H1-EGFP-doxorubicin FRET
To estimate the intracellular DOX concentrations, we have fitted the H1-EGFP emission decay profiles to a double exponential decay model that assumes a single FRET efficiency, i.e. that there is a longer lifetime fluorescence decay component representing the non-FRETing H1-EGFP emission and a single shorter lifetime component representing the FRETing H1-EGFP emission. This approach enables us to obtain reasonable decay χ 2 goodness of fit values with only a single adjustable fit parameter per pixel. When DOX binds to DNA, however, it can have a range of distances relative to the nearest EGFP-tagged H1 protein and there can be a range of angles between the DOX and EGFP fluorophore dipole orientations, so there will be a distribution of Förster resonant energy transfer efficiencies between EGFP and DOX. However, similar considerations apply to many FRET measurements of populations of fluorophores and it is common practice to analyse FLIM FRET data in this way. A further consideration is that the EGFP fluorophore is effectively static during its fluorescence decay and so the conventional assumption that FRET measurements average over a rapidly varying set of random orientations between the donor and acceptor fluorophore dipoles is not valid. This consideration is relevant to all FRET data measured with fluorescent proteins and means that the donor fluorescence will not be fully represented by a discrete exponential decay model 1 . Here, however, because we apply the same FLIM/FRET analysis to both the in vitro calibration data and the in vivo data, any bias due to the fitting model used should cancel as we apply our look-up table correction to obtain the estimated equivalent in vitro DOX concentrations.
Our simulation above shows that the DOX fluorescence accounts for ~1% of the total fluorescence signal for an in vitro DOX concentration of 1 µM (Supplementary Figure 3b). This is supported by direct comparison of in vitro LSM and CEM data, which shows good agreement in the measured non-FRETing H1-EGFP fractions up to a DOX contribution of ~2 µM (Supplementary Figure 3c). Above this DOX concentration, the effect of DOX fluorescence on the measured decay becomes more significant and we use the correction procedure based on a look-up table derived from Supplementary Figure 3c to convert the measured in vivo fraction of FRETing H1-GFP to the equivalent in vitro DOX concentration. While this correction is valid provided that the bleed-through contribution of DOX to the detected fluorescence signal is the same in vivo and in vitro for a given DOX concentration, there are a number of factors that may affect the validity of this assumption. First, DOX fluorescence intensity has been previously shown to be sensitive to its bound state 2,3 , which may be different in vivo compared to in vitro. To reduce the potential effect of this on our analysis, we have used nuclear image segmentation to exclude potential contributions from unbound cytoplasmic DOX. We also note that DOX has previously been shown to have an intracellular degradation product with a longer fluorescence lifetime 4 . Again, any effect from this will be greatly reduced by the nuclear image segmentation applied because this degradation product was observed to accumulate in the cytoplasm. A further issue is that the fluorescence lifetime of DOX has been shown to be sensitive to the degree of chromatin condensation [5], which also may be different in vivo compared to in vitro. However, as the range of change in DOX lifetime with cell state varies by less than a factor of 2 5 and, given that the unwanted DOX signal in our CEM data is a small fraction of the total fluorescence, we believe that this effect will not greatly perturb our results, particularly for lower DOX concentrations.