All results were obtained in living cancer cells as confirmed by viability tests, which showed that around 70% of both treated cells (TC) and not treated cells (NTC) were still viable (i.e. alive) even after the neutron scattering experiments that were performed at room temperature for several hours. First, optical microscopy analysis was performed to observe morphological changes in human MCF-7 breast cancer cells due to PTX’s action. It is known that PTX’s action is driven by its interaction with microtubules and the consequent arrest of cells in mitosis12. As a consequence of this process, the rounded cell morphology observed in the TC is expected, as shown by the representative phase contrast microscopy images of the studied cells in Fig. 2.

Figure 2


Representative optical microscopy images of breast cancer cells (MCF-7) not treated (NTC – red circle) and treated (TC – blue circle) with 15 nM of paclitaxel for 24 h. The rounded cell morphology observed in the TC is expected due to PTX’s action. The PTX molecule is shown in the left corner, where blue represents the carbon atoms, red represents nitrogen, cyan represents oxygen and the H atoms are shown in white.

In seeking to investigate how these morphological changes due to PTX action cause perturbations within cellular water, the cells were studied by means of differential scanning calorimetry (DSC) at temperatures around the water melting point. The results are given in Fig. 3(a,b), together with pure water (bulk) that was also analyzed for better comparison. The accessibility and fastness of this experiment allowed us to perform this analysis in duplicate using cell cultures independently prepared. Specifics are presented as Supplementary material.

Figure 3


Differential scanning calorimetry (DSC) performed on bulk water (black circles) and on breast cancer cells (MCF-7) not treated (NTC – red circles) and treated (TC – blue circles) with 15 nM of paclitaxel for 24 h. (a) Respective melting peaks and (b) specific heats (baselines) before and after melting. The samples masses were approximately 25 mg.

Water, as a pure substance, has its melting point determined by the onset on the curve, while for water in the cells the melting point is determined by the peak position of the respective transitions13. As a result of a distribution of confined water populations, both TC and NTC show melting peaks with tails at the low temperature side. At the high temperature side, the cells exhibit melting rates similar to that observed for water, thus depicting the presence of a bulk-like water population. Regarding the NTC, the melting peak is centered at 0.8 °C, indicating a melting temperature close to that observed for water. Meanwhile, the melting peak of the TC is shifted to −1.8 °C and shows a 5% reduction of the area under the curve. This points to a reduction in the melting enthalpy, from 202 J/g for NTC to 192 J/g for TC (while for pure water we have determined the melting enthalpy as 367 J/g), thus showing that in the TC, the ice structure is somewhat less stable and suggesting a slightly higher content of confined water. After re-scaling the bulk water data to match the melting peaks from each cell’s data, we observe that bulk water represents around 60% of the signal in both cases. While not purely quantitative, this simple method provides an approximate amount of bulk water in the cells, which is within the range recently determined for distinct living cells14. Graphical representations of this analysis are also presented as Supplementary material.

Focusing now on the analysis of the baseline of each sample, as highlighted in Fig. 3(b), information about the specific heat of the NTC and TC can be obtained and compared before and after melting (solid and liquid phases, respectively). In the solid state, the TC likely have higher specific heat values, indicating an increase of phonon vibrations15 of the ice structure. Because of the very broad range of the melting transitions further confirmation of this assumption was needed. This was achieved by INS as described below. At biological temperatures, i.e. above melting, despite the oscillations in the data, the TC (blue line) shows lower specific heat, and thus stronger response to temperature changes.

To deepen our understanding on the changes in the cellular water behavior before melting, we turn to the INS experiments depicted in Fig. 4. INS data was obtained in a cumulative way for 12 hours at 10 K using the TOSCA spectrometer located at the ISIS Neutron and Muon Source, UK, on both types of cells, pure (bulk) water and pure PTX. Figure 4(a) presents the spectra for TC, NTC and bulk water. The latter has been re-scaled for better visualization, while the PTX spectrum, shown in Fig. 4(b), was scaled to reflect the drug concentration of 15 nM added to the TC. Particular regions of the cells’ spectra can be well described as follows. Contributions from amide groups from proteins and DNA16 are assigned to vibrations above 140 meV (1120 cm−1), while the spectral region between 50 and 130 meV (403 and 1048 cm−1) is mainly ascribed to the water librational modes, where the bulk-like population dominates the spectra. Between 20 and 45 meV (161 and 363 cm−1), the sharp peaks are assigned to stretching of water H-bonds, while the broad ones observed in the cell spectra around 32 and 25 meV (258 and 202 cm−1) result from torsional motion of CH3 groups in proteins and DNA16. Finally, the spectral region between 0 and 20 meV (161 cm−1) is attributed to lattice vibrations, i.e. phonon modes, of both water and proteins. This spectral region exhibits the clearest difference between NTC and TC spectra.

Figure 4


(a) Inelastic neutron scattering (INS) spectra for breast cancer cells (MCF-7) not treated (red) and treated (blue) with 15 nM of paclitaxel for 24 h and for bulk water (ice Ih form) (black). Data was collected on one sample of each cell group. The volumes of the samples were 2.5 mL and consisted of 3 × 107 cells/mL. The presented results were obtained after 12 h of data collection for each sample allowing for an optimal fractional error bar as defined by Poisson statistics. (b) Contribution from the confined water of TC and NTC cells obtained after subtraction of bulk water contribution. Also in (b), the green curve shows INS data for PTX, which was scaled to reflect the drug concentration of 15 nM added to the TC. PTX spectrum is presented as Supplementary information within a scale that allows for observing the molecule vibrational modes together with their assignments.

To gain further insight solely on the confined water contribution in the NTC and the TC, we have subtracted the bulk water contribution from both spectra, using the information obtained from the DSC analysis (60% of the water content in the cells have bulk-like behavior). For that, prior to data subtraction, we have re-scaled the INS data of bulk water between 50 and 130 meV (403 and 1048 cm−1), where all the contributions can be assigned to water molecules. Thus, the plots shown in Fig. 4(b) predominantly represent the contribution from confined water together with the remaining cell components. Following this approach, the subtle difference between the NTC and TC spectra at the lattice vibrations spectral region, 0 and 20 meV (161 cm−1), is further highlighted. This observation strengthens the calorimetric results that have already indicated that the action of the drug modifies the way in which water molecules are structurally arranged in MCF-7, validating the differences observed in the specific heat in the solid state. On the other hand, vibrations between 20 and 45 meV (161 and 363 cm−1), attributed to proteins and DNA molecules, are similar for both samples. Since the amplitude of vibrational bands are directly dependent on the concentration of vibrating species, this result shows that changes in confined water properties are not related to changes in proteins concentration within the MCF-7 samples.

Although the thermal analyses insights for the cells in the solid state can be explained based on changes of the lattice vibrations, as shown by INS, the same is not true for the cells above the melting point. In this case, following the description proposed by Bolmatov et al.17 to describe the thermal properties of liquid systems, it is expected that diffusional and rotational motions play an important role in the specific heat of the cells. Differences in the water dynamics above melting, already indicated by slight differences in the specific heat, are expected to be even more evident in the QENS data obtained at ambient temperature using the BASIS spectrometer located at the Spallation Neutron Source, Oak Ridge National Laboratory, USA. In these experiments, the samples were also measured for approximately 12 h for reduction of fractional error and optimization of signal-to-noise ratio. Here it is worth to mention that data collected during the first minutes were compared to the cumulative data and no changes neither in the INS nor in the QENS spectra were observed. This demonstrates the integrity of the samples as well as the reproducibility of the data.

In a QENS spectrum, dynamic components typically manifest themselves as Lorentzian curves, whose line-widths, Γ, behavior as a function of the scattering wave-vector, Q, allows differentiating diffusive and localized motions. While rotation of molecules do not show any clear Q-dependence in the QENS signal width, the same is not true for translational motions18. Also, by representing the QENS as the imaginary part of the susceptibility19,20:

$$\chi ^{\prime\prime} (Q,E)\propto \frac{S(Q,E)}{{n}_{B}(T,E)}=S(Q,E)\ast [exp(\frac{E}{kT})-1]$$


where nB(T, E) is the temperature Bose factor, different relaxation processes can be clearly separated; changes in peaks position, shape and/or intensity are good indications for Q-dependent processes, i.e. water diffusion.

Because the motions of purely bulk water molecules are usually too broad (fast) to be discernable from the background due to the instrumental design of BASIS, any bulk-like behavior observed in this descriptive study is likely attributable to water inside the cells, rather than extracellular water, whose dynamic properties are expected to be much closer to pure bulk behavior20.

Figure 5(a,b) show respectively the imaginary parts of the susceptibility for selected Q-values for the NTC and TC. While for the NTC relatively little Q-dependence is detected, the opposite is observed for the TC. After being treated with PTX, the MCF-7 cells present a Q-dependent relaxation, centered around 10 μeV (2.4 GHz) at Q = 0.3 Å−1 and is gradually shifted to higher frequencies as Q increases. For these cells, we can therefore infer the presence of non-localized (diffusive) motions on the time and length scales assessed by the instrument.

Figure 5


Dynamic susceptibilities obtained from QENS data for not treated breast cancer cells (MCF-7), NTC (a) and breast cancer cells treated with 15 nM of paclitaxel for 24 h, TC (b). The presented results were obtained after approximately 12 h of data collection for each sample allowing for an optimal fractional error bar as defined by Poisson statistics. Data was collected on one sample of each cell group. The volumes of the samples were 2.5 mL and consisted of 3 × 107 cells/mL.

The presence of diffusive motions in the TC is further confirmed and quantitatively discussed by the analysis of the evolution of the QENS broadenings (half width at half-maximum, HWHM) vs Q2 presented in Fig. 6. Here, the HWHM values were obtained by performing a nonlinear least square fitting to the QENS signal at different Q-values. A single Lorentzian function, convolved with the instrumental resolution was used as input model together with a baseline correction that accounted for the background. The χ2 (chi-squared) test was used as the statistical analysis tool for goodness-of-fit. The results are presented as Supplementary material together with the graphical representations of the fits. For comparison purposes, Fig. 6 also shows a theoretical curve depicting the expected behavior for purely bulk water at room temperature (black dotted line), obtained by considering the jump diffusion model, described by the expression below21:

$$HWHM(Q)=\frac{D{Q}^{2}}{1+D{Q}^{2}{\tau }_{0}}$$


where D refers to the diffusion coefficient and τ0 is the residence time. For bulk water we have used D = 2.9 × 10−9m2/s and τ0 = 1ps22.

Figure 6


Evolution of QENS broadening (half width at half maximum – HWHM), determined from fits (see Experimental section and supplementary material) and attributed to the water molecule motion in breast cancer cells (MCF-7), as a function of Q2. The symbols represent the values obtained when the HWHM was allowed to vary independently at each Q-values, while the error bars represent the standard deviation. Note that some of the error bars are within the dots. The black dotted line represents a theoretical curve for bulk water obtained from the jump diffusion model, Equation (2). The same model was used to obtain the blue dotted curve, which describes the behavior of the TC. The red dotted line represents a linear fit that describes the HWHM behavior in the NTC. The data collected at Q2 = 2.25 Å−2 were not used due to the presence of a Bragg reflection. In the bottom right corner, the scheme illustrates the appearance of translational motions in the cells after treatment with PTX.

For TC, the QENS broadening presents a clear Q-dependence below 0.7 Å−1, confirming the presence of non-localized, long-range translational motions. Above this value, the motions become too fast for the instrumental range of accessible energy transfers and fitted values of the measured signals gradually decrease. On the other hand, for the NTC, the HWHM is Q-independent, indicating spatially localized (rotation motions) and short-range relaxations. These observations quantify the conclusions drawn from the model-independent susceptibility data, thus offering clear evidence for a less spatially constrained mobility of the intracellular water molecules in the TC when compared to that in the NTC. The proposed change in dynamics behavior before (red) and after treatment (blue) with PTX is illustrated in the bottom right corner of Fig. 6.

Sequentially, the curve obtained for the expected behavior of bulk water via Equation (2) was adjusted to the data from the TC below 0.7 Å−1 by a nonlinear fitting procedure. Hence, the curve describing the water dynamics in the TC was obtained with DTC = 2.21 ± 0.11 × 10−9 m2/s and τ0-TC = 9.09 ± 0.62 ps. While the diffusion coefficient, DTC, is comparable to the one expected for bulk water, the significantly higher residence time shows that the H-bond network in this water population is considerably disturbed. The same is true for the NTC, where a rotational relaxation time of about 67 ± 3 ps can be extracted (\({\tau }_{{rotation}}\,\approx \,2\hslash /{{\rm{\Gamma }}}_{r}\)23) by a linear fitting as presented in Fig. 6 by the red dotted line. In both cases, the error bar depicts the standard deviation of the HWHM value determined by the fit. The comparison clearly demonstrates that the water structural network in the TC differs from the one present in the NTC and that there is higher mobility freedom for the water in the TC.