Label-free DNA imaging in vivo with stimulated Raman scattering microscopy

Noting that SRS and spontaneous Raman spectra are mostly identical (24), the feasibility of imaging DNA in the CH band is evident from the distinct Raman spectra of DNA, protein, and lipid. In the CH stretching region, the DNA peak frequency (2,956 cm⁻¹) is higher than those of protein (2,931 cm⁻¹) and lipid (2,854 cm⁻¹) (Figs. S1A and S2F). These assignments were confirmed with the Raman spectra of cellular extractions (Fig. 1B). Comparing the Raman spectra of deoxyribose phosphate, adenine, thymine, guanine, and cytosine (Fig. S2 A–E), we realized that the DNA’s CH vibrational signal comes primarily from the deoxyribose. We attributed the blue shift of DNA’s CH peak from the protein and lipid peaks to the fact that there are more carbon atoms in deoxyribose than are in proteins and lipids bonded to oxygen atoms, which are more electronegative.

We tested SRS imaging with an artificial sample composed of DNA fibers, BSA as the protein source, immersed in a drop of oleic acid (OA) as the lipid component. For these three components, we acquired three SRS images at three Raman shifts to map the distribution of each compound by linear decomposition. The optimal Raman shifts were selected by theoretical calculation based on the Raman spectra (Materials and Methods). SRS images were automatically acquired in tandem through synchronizing the tuning of the laser wavelength to the frame trigger of the microscope. Next, linear decomposition was performed using a premeasured calibration matrix. The distribution of the DNA fiber, protein crystal, and oleic acid were mapped and the three compounds were spatially separated with the expected contents on prior knowledge of the sample (Fig. S1 B–E). Thus, this experiment has provided a proof-of-principle to our method (linear decomposition of multicolor SRS imaging). This approach avoids full spectra data acquisition and statistical analysis, offering rapid SRS imaging with multiple chemical contrasts and relatively lower photodamage.

In addition to these major components of interest, the existence of other unknown or ignored components may reduce the accuracy of the linear decomposition. However, when their signal contribution to the total signal is little, then they can be considered to be negligible. Linear fitting of the Raman spectra of DNA, cellular proteins, and lipids (Fig. 1B) to the spectrum of a cell pellet, suggested that these three major compounds account for ∼90% of the total of the CH stretching vibration of the cells (Fig. 1C and SI Text), indicating that the overall accuracy of our approach is remarkable and could be used to study biological samples.

We proceeded with simultaneous imaging of DNA, protein, and lipid in live cells (Materials and Methods and Fig. S3A). We started from imaging cells in mitotic phase (prophase) with condensed chromosomes undergoing cell division. SRS images at three Raman shifts (2,967, 2,926, and 2,850 cm⁻¹) were acquired following the approach described above. After linear decomposition, the distribution of nucleic acids, proteins, and lipids were mapped (Fig. 1D). Chromosomal DNAs were visualized based on nucleic acids contrast with a high sensitivity. By the optical sectioning capability of SRS, we reconstructed a 3D codistribution of DNA and lipids in a single cell (Fig. S4A), showing that, although lipids are widely distributed throughout the cell cytoplasm, their density within the cell nucleus is very low, as expected. Protein distribution appears to be more uniform throughout the entire cell.

We next imaged interphase cells and were able to obtain DNA contrast with a high signal-to-noise ratio (Fig. 1E). Chromatin structures within the cell nuclei such as nucleoli, heterochromatin, and euchromatin were visualized at the optical diffraction-limited resolution, demonstrating that SRS imaging of DNA in the CH band offers higher sensitivity than that of the fingerprint spectral region (22).

We were also able to capture chromosome dynamics during in vitro cell division by time-lapse SRS imaging. The chromosomes in a metaphase cell are organized along a line in the center of the cell and then are equally split into two parts as the cell enters in anaphase (Fig. 1A, Fig. S3A, and Movie S1). SRS imaging captured these different chromosomal dynamics without inducing obvious photodamage to the live cells, as evidenced by the fact that metaphase cells could finish their natural division process with successful passage through the critical M-checkpoint. This experiment demonstrates that label-free live cell imaging with SRS is a powerful method to examine chromosome dynamics during cell division. With this approach, it is also possible to observe nucleolus disassembly and reassembly, to determine the condensation level of DNA, and to monitor the movement of the chromosome during cell division, as well as other crucial cellular processes (25).

Noteworthy, we observed that there is minimal cross-talk among the three decomposed images of DNA, protein, and lipid, demonstrating a good decomposition of molecular compounds. The cell nuclei are clearly visualized with a positive contrast from the DNAs, whereas the ribosomal RNAs are mainly distributed in the cytoplasm, contributing only as a weak background (Fig. 1 D and E and SI Text). To verify the SRS chemical contrast for DNA, we first imaged the cells with SRS and then labeled the cells with a DNA fluorescent dye (DRAQ5) for two-photon excited fluorescence (TPEF) imaging, showing that the paired images matched well with each other (Fig. 1F).

We conducted in vivo SRS imaging of DNA in mouse skin to follow cell division activity. 12-0-Tetradecanoylphorbol-13-acetate (TPA) is a potent tumor promoter that has been used to induce epidermal carcinogenesis. Topical treatment of mouse skin with TPA induces epidermal hyperplasia, characterized by significant increase in skin thickness and mass, total number of the cell nuclei, and mitotic rate (26), thus offering an ideal model to study chromosomal dynamics during cell division. Although the proliferative effect of TPA on the mouse skin has been reported, detailed cell cycle kinetics of this process has not yet been well studied due to the lack of proper tools to visualize DNA in vivo.

We therefore performed an in vivo imaging of TPA-treated mouse skin using a skinfold chamber model (Materials and Methods and Fig. S3 B and C). We found that the number of mitotic cells was significantly increased in the epidermis (yellow arrowheads in Fig. 2A) with respect to their control counterparts (Fig. 2B). This mitotic cell state was easily identified based on stronger DNA signals (Fig. 2D), as well as its distinctive morphology of condensed chromosomes, compared with a cell nucleus in interphase (Fig. S4B). Fig. 2C shows representative images of cell nuclei at different stages of a complete cell cycle, in which nuclear morphology, including internal detailed structures, was clearly visualized. In addition, unlike in vitro cultured cells, very few lipid droplets were observed in live mouse skin tissue. Fluorescent staining was used to confirm the DNA contrast of SRS in intact fresh skin tissue (Fig. S4C). We also captured the dynamic of a cancer cell during division by recording an in vivo SRS movie with the same skinfold chamber on an immune-deficient mice injected with human cancer cells to create a xenograft model (Materials and Methods). The splitting of the chromosomes was easily seen when the cell was entering anaphase, by using time-lapse SRS imaging based on DNA contrast (Movie S2).

We then tracked the cell cycle kinetics during the proliferation process through performing in vivo mitotic counting in TPA-treated mouse skin, by applying time-lapse SRS imaging based on DNA contrast (Materials and Methods and Figs. S3D and S5). Fig. 2E shows the mitotic rates (number of mitotic cells per thousand cells) over a 24-h period with a 6-h interval. Our data show that mitotic activity reached a peak at ∼18 h and then decreased at ∼24 h (Figs. S6–S9). This result confirmed that a synchronized wave of basal cell proliferation is induced by TPA in adult mouse skin. We noted that in vivo SRS imaging of DNA makes this type of dynamic studies possible because of its unique proficiencies, including label-free intrinsic chemical contrast, high sensitivity, and 3D sectioning capability, with no photo bleaching.

In addition, SRS offers the possibility of label-free imaging of live cancer cells from primary tumors, which is difficult to label with fluorescence without perturbing cell functions. We anticipate important clinical applications of this approach, such as assessing and screening morphologic effects of antineoplastic agents in real time (27). For in vivo imaging of live animals, a modified skinfold chamber model was used, and a subskin implantation model was developed to minimize the breathing motion artifacts in our experiments, which could also be applied for other microscopic imaging modalities. Future SRS technical developments, such as simultaneous multicolor imaging, video rate scanning, and the design of endoscopic or handheld probes, will further release the critical requirements to the animal immobilization strategies.

We broadened the use of SRS to image human skin tissue. The nuclear morphology in normal human skin was clearly visualized with positive DNA contrast with a high sensitivity, as well as images of protein and lipids, which displayed the regularly layered skin structures (Fig. 3 A and B), providing complementary tissue and cellular morphological information. To first validate the DNA contrast in human tissue, we imaged the same skin tissue section with SRS and H&E staining in tandem. We found a clear correlation in terms of visibility of detailed nuclear morphology and architecture, as well as cellular and tissue morphology (Fig. 3 C and D), confirming the effectiveness of our method.

To demonstrate the potential of SRS for label-free histology of cancer in humans, we imaged fresh human skin cancer tissue from three surgical cases of squamous cell carcinoma (SCC), the second most common type of skin cancer (28). We found that we could easily identify an increased number of mitotic figures based on stronger signals and distinctive morphological features of the condensed chromosomes (yellow arrowheads in Fig. 3E) (29). Mitotic figures are valuable diagnostic and prognostic indicators of cancer aggressiveness (30), becausew it correlates directly with the level of cell division and proliferation. Fig. 3F shows another representative image of a small nest of carcinoma cells, in which aggregated tumor cells with enlarged nuclei (right side of the dotted curve) are surrounded by nonneoplastic cells with smaller nuclei (left side of the curve), reflecting high intratumoral heterogeneity (31). Our results demonstrate that the multicolor SRS approach for label-free imaging of DNA, protein, and lipids in tissues offers clear and equivalent histological features as conventional H&E staining does for skin cancer diagnosis, with the advantage of being a label-free method and thus not affecting the native form of the tissue. Although other multiphoton imaging techniques such as native TPEF and second harmonic generation (SHG) can also reveal most of the tissue morphological features (32, 33), SRS provides chemical specificity for nucleic acids. SRS therefore highlights both the nuclear morphology and also allows for quantification, enabling identification of mitoses and nuclear atypia in a quantitative fashion. We expect that SRS histology may not only speed up Mohs surgery by on-site label-free imaging of tumor tissue with margins, but also has the potential for in vivo noninvasive detection and progress evaluation of skin lesions in real time.

We used the picoEMERALD laser source (APE), which comprises an optical parametric oscillator (OPO) synchronously pumped by a frequency-doubled picosecond oscillator (High-Q Laser) in a single housing. The OPO supplies the pump beam (5–6 ps, tunable from 720 to 990 nm), and the oscillator supplies the Stokes beam (7 ps, 1,064 nm). The two beams are temporally synchronized and spatially overlapped and then are coupled into a modified laser-scanning confocal microscope (FV300; Olympus) for SRS imaging. This picosecond system maps the sample of a single Raman shift at a time. To do spectral or multicolor imaging, the wavelength of the pump beam is scanned by tuning the Lyot filter in the OPO cavity. In our experiment, we synchronized the tuning of the Lyot filter to the frame trigger of the microscope through the RS232 serial port by Labview programming to realize automatic image acquisition at optimally selected multiple Raman shifts frame by frame, which made our multicolor SRS microscope feasible for long-term time-lapse imaging of live cells and live animals in vivo. Each frame (512 × 512) was taken recurrently within 1 s to a few seconds. We used a high NA water immersion objective lens for imaging (UPlanApo IR 60× NA 1.2; Olympus).

We used an artificial sample to demonstrate the multicolor approach with linear decomposition. The sample was composed of DNA fibers (Sigma) and a piece of BSA crystal (representing protein; Sigma), immersed in a droplet of oleic acid (OA, representing lipid; Sigma). Mathematically, for three components, at least three images should be acquired at three Raman shifts. The Raman spectra of DNA, BSA, and OA in the high wavenumber range of the carbon-hydrogen (CH) stretching vibrational band (2,800–3,050 cm⁻¹) are shown in Fig. S1A. Although they largely overlap, they clearly do show distinct spectral features. As the spectra of SRS and Raman are mostly identical, to select the most distinct spectral features to decompose the three components (DNA, BSA, and OA) with as much accuracy as possible, we performed a simulation based on the criterion that minimizing the root mean square error (RMSE) between the concentrations of the mixture components and their estimates to select the optimal Raman shifts (34–36).

For a sample containing M components with unknown concentrations {c_m| m = 1…M}, respectively, we measure SRS signals at N wavelengths {ω_n| n = 1…N} (N ≥ M). The optimal wavelengths can be determined by minimizing the RMSE between the true concentration and the calculated concentration

where tr, E, and t denote matrix trace, expectation, and transpose, respectively. The measured spectrum {S_n} is equal to the linear combination of the spectra of the M components {k_m(ω_n)} weighted by the concentrations and the detection noise {r(ω_n)}

where

and

The least square solution for Eq. 2 is

Substituting Eqs. 2 and 3 to Eq. 1 yields

Given that the detection noise in SRS imaging is near shot-noise limited and obeys an uncorrelated process with zero mean and a constant variance σ², Eq. 4 reduces to

where the calibration matrix K was based on the Raman spectra of the pure chemicals. Our simulations using Eqs. 1 and 5 allowed the selection of three optimal Raman shifts at 2,973, 2,921, and 2,851 cm⁻¹ (Fig. S1).

The Raman spectra were acquired using a confocal Raman spectrometer (LabRAM HR800; Horiba Jobin Yvon). A Helium-Neon (HeNe) laser at 633 nm was used to excite the sample. The spectra were processed using LabSpec software. The objective is RMS20X (Olympus: NA, 0.4, WD, 1.2 mm). The average power on the sample was ∼20 mW. Integration time for each spectra was ∼10–30 s.

TPEF imaging of DNA stained with DRAQ5 (BioStatus) in live cells and tissue was excited by the 1,064-nm beam (by blocking the OPO beam) and detected by a photomultiplier tube (PMT; Hamamatsu) through a dichroic mirror and a short-pass filter. The samples were first imaged with SRS and then were stained with DRAQ5 (10 μM, 10 min for live cells and 30 min for fresh tissue at room temperature) directly on the microscopy stage (37).

HeLa S3 cells (ATCC) were maintained at 37 °C in a humidified 5% (vol/vol) CO₂ air incubator and cultured in MEM (Invitrogen) supplemented with 10% (vol/vol) FBS and 0.01 mg/mL insulin (Sigma). Cells were imaged in phenol red-free MEM (Sigma) supplemented with sodium bicarbonate, 20 mM l-glutamine,100 ng/mL epidermal growth factor (EGF; Sigma), and 25 mM Hepes, pH 8.0 (Invitrogen).

Time-lapse imaging of live cells for observing the cell division process was carried out on a 35-mm dish with a 0.17-mm coverglass bottom (MatTek) in a modified onstage incubator (Live Cell Instrument; Chamlide). To accommodate both high NA objective and high NA condenser with very short working distance (<1 mm) for SRS imaging, the upper cover of the incubator was specifically designed using flexible plastic materials to enclose the objective within the incubator and to allow it to move for aligning and focusing at the same time (Fig. S3A). The temperature was maintained at 37 °C with 5% (vol/vol) CO₂ in humidified air. An extra 35-mm dish with 1× PBS solution was placed within the incubator for better humidity maintenance.

For in vitro cell division imaging, HeLa cells were synchronized by double thymidine block (38). This procedure involved 18-h incubation (prepared at ∼25–30% confluence) in thymidine (final concentration, 2 mM; Sigma), 9-h incubation in fresh media, and again a 17-h incubation in thymidine (final concentration, 2 mM). Before imaging, cells were incubated in fresh media for 8 h to allow cells reach middle G2 phase. For in vivo cell division imaging in the xenograft mouse model, HeLa cells were also synchronized using double thymidine block. After releasing, incubated cells in fresh media for 3 h before injecting into the mouse skinfold chamber. The total cellular lipids from the cultured HeLa cell pellet were extracted using the Bligh-Dyer method (39). Total cellular protein was extracted using RIPA buffer (Thermo Scientific). We did not extracted DNA from the cell pellet.

To minimize the motion artifacts in live animal imaging with multicolor SRS, we built the dorsal skinfold chamber model on mice. The skinfold chamber, which consists of two symmetrical titanium frames, was implanted in the dorsal skin of the mice following previously reported procedures (40–42). The chamber sandwiched and immobilized the skin on the back of the mouse. One side of the skin was surgically removed and replaced by a round 0.17-mm coverglass with a 10-mm diameter. The screws used should be as short as possible on both sides of the chamber to allow the high NA objective and the high NA condenser to be able to access to the chamber as close as needed for SRS transmission imaging (Fig. S3 B and C). We used this model for two experiments as follows. (i) For the experiment of in vivo mouse skin imaging, young adult mice (Swiss Webster, female, 6 wk; Taconic) were used. The skin within the chamber was topically treated with 2 µg TPA (Sigma) in 200 µL acetone to activate cell division activity (26, 43, 44). (ii) For in vivo cell division dynamic imaging experiments, Ncr nude female mice (Taconic) were used to establish a xenograft model (45, 46). Synchronized HeLa cells with double thymidine block were injected into the chamber superficially beneath the inner side of the skin. Images were taken after 26 h of the injection.

Mouse dorsal skin was topically treated with 8 µg TPA in 800 µL acetone to activate cell division activity in the skin. To immobilize the skin for imaging, a thin metal piece was surgically implanted under the skin (titanium, 12 × 12 × 1 mm) with two small handles (2 × 5 × 1 mm) extended out of the skin. When imaging, the two small handles were fixed to a mounting block on the microscope stage with the mouse under anesthesia. After imaging, the mouse was released and maintained in the cage in normal conditions. All of the in vivo animal experiments were performed under standard anesthesia condition using O₂/isoflurane (Fig. S3D). Animal experiments were conducted in accordance with Harvard University IACUC Protocols 10-02 and 29-01.

Discarded and deidentified human skin samples including normal (from five surgical cases) and SCC (from three surgical cases) tissue were collected and prepared in accordance with Massachusetts General Hospital institutional review board protocol 2013-P-2337. To collect paired images with SRS and H&E from the same tissue slice, fresh normal skin tissue was quickly embedded and frozen in OCT compound for frozen sectioning without fixation; 20-μm slices were first imaged with SRS and then were stained with H&E (following standard procedures) for light microcopy imaging. Fresh SCC tissue was cut into thin slices using a blade (∼1 mm) and then was sandwiched between a slide and coverglass for SRS transmission imaging. Some pressure was applied to flatten the tissue.

Raman spectroscopy has been used to characterize the biochemical composition of cells and tissue (47–55). Raman imaging of DNA (cell nuclei) in single living cells is often obtained through full spectral decomposition or statistical data analysis (23). Because spontaneous Raman signals are extremely weak, the dwell time for each pixel is very long (70–80 ms), which means that, for example, in a high-resolution image of 512 × 512 pixels, using a high NA objective lens, the imaging time could be up to a few hours (23). In contrast, our approach of multicolor SRS with linear decomposition takes only 10–20 s to retrieve the DNA contrast with both high sensitivity and high spatial resolution, which is significantly faster (∼1,000 times) than spontaneous Raman imaging. Furthermore, with technical improvements, such as simultaneous three-color imaging, to replace sequentially tuning of the laser wavelengths, as well as more sensitive detectors, it is possible to further reduce the imaging time.

DNA imaging based on individual Raman peaks (i.e., 785 cm⁻¹ for symmetric phosphodiester stretch ring breathing modes of pyrimidine bases of nucleic acids and 1,090–1,010 cm⁻¹ for symmetric dioxy stretch of the phosphate backbone) in the fingerprint spectral region (400–1,800 cm⁻¹) has also been reported with both spontaneous Raman (56) and SRS (17). However, thus far, high-quality DNA imaging has only been achieved for cells with highly condensed chromatin/chromosomes, such as polyploid cells, mitotic cells, apoptotic cells, and sperm (17, 57, 58), but not for interphase cells.

Single-color (single-frequency) SRS microscopy maps the distribution of a single Raman peak with high chemical specificity. However, because many biomolecular vibrational modes usually overlap, single-color imaging is incapable of differentiating a specific chemical compound from a mixture, as would be the case in biological tissues (16). To decompose the mixture, spectral imaging with linear fitting or statistical analysis has to be performed. However, for imaging of dynamic samples, such as live cells and live animals, a rapid and simple imaging approach is critical. Because there may exist redundant information in the full spectral data, it is reasonable to imagine that only a few optimally selected Raman shifts, to roughly retrieve individual chemical distributions, would be sufficient. Mathematically, to distinguish M components, at least M-color images should be acquired (21).

Noteworthy, in a sample with M major components of interest and other ignored or unknown components (I) as an interfering background, the latter will decrease the accuracy of the linear decomposition. When these I components contribute little (i.e., <10% of the total signal), they are considered to be negligible, and the overall accuracy for the sample can be based only on the M-component analysis and be still quantitative/semiquantitative. In our experiment, only the major biomolecules containing abundant CH bonds or groups were considered (nucleic acids, protein, and lipids), whereas others with relatively lower signal contribution were ignored. Water, for example, was ignored because its signal was relatively small at the three Raman shifts we used within the CH stretching vibrational band and also because its distribution was rather uniform in all cells and tissues. Because of the relatively more uniform distribution of RNA than DNA, we were justified in presenting localized features as being due to the DNA. In addition, the DNA contrast achieved with this three-color approach presented variations in different tissue types, and therefore, the imaging strategy had to be optimized for each tissue type. Recent technical developments in SRS microscopy, such as hyperspectral detection in parallel or with fast wavelength tuning, is making the rapid collection of the full spectral data to be more and more feasible (59, 60).