2.1.

Imaging System

The light source used is an ultrafast laser operating at $900\mathrm{nm}$ with $100\text{-}\mathrm{fs}$ pulse duration (Mai Tai Titanium Sapphire Laser, Spectraphysics, Newport Corporation, Irvine, California). The imaging was performed using a confocal microscope (LSM 510 Meta, Carl Zeiss GmbH, Jena, Germany). A source laser passes through the prism-based pulse compressor (Femto Control, APE GmbH, Berlin Germany) followed by the acousto-optic modulator (AOM), and is then focused on the sample with a $20\times$ objective lens with numerical aperture $\left(\mathrm{NA}\right)=0.5$ . The SHG signal was measured on the other side of the sample, filtered through a $450\text{-}\mathrm{nm}$ bandpass (BP) filter (full-width half- $\text{maximum}=10\mathrm{nm}$ ) before reaching the photomultiplier tube (PMT) (R6357, Hamamatsu Photonics, Hamamatsu City, Japan). The schematic of the microscope setup is shown in Fig. 1 .

Fig. 1

Schematic of the PM-SHIM setup. The setup of the PM-SHIM is shown on the left and the alignment configuration for chirp analysis is shown in the right. The PM-SHIM consists of a femtosecond (fs) laser followed by the pulse compressor, AOM, and the confocal microscope setup. The AOM is used to modulate the power delivered to the sample using a diffraction grating. The laser beam reaches the sample, and the SHG and TPEF signals from the samples pass through a bandpass (BP) and short-pass (SP) filter in transmission and reflection modes, respectively. The signals are collected using a photomultiplier tube (PMT). (a), (b), and (c) are three positions at which the laser beam is characterized using the chirp analyzer. The laser beam from the femtosecond laser, (a) before the AOM, (b) after the AOM, and (c) at the sample stage is guided to the chirp analyzer using mirrors (M1 to M6) and a beamsplitter. The beamsplitter (BS) is used to reduce the laser power by 50% in (a) and to split the beams orthogonally in (c), and then is used to guide by means of mirrors to the chirp analyzer.

A chirp analyzer (GRating-Eliminated No-nonsense Observation of Ultrafast Incident Laser Light E-fields, Swamp Optics, Atlanta, Georgia) was used for measuring the dispersion profile of the laser resulting from pulse compression.³⁷ The beam profile was measured at three locations: 1. before the AOM, 2. after the AOM, and 3. at the sample stage [see Fig. 1]. The optical setups for all the measurements are shown in Fig. 1. For all positions, pulse duration was optimized by systematically adjusting the prism positions inside the pulse compressor.

2.2.

Xenograft Model, Tissue Isolation, and Collagen Hydrogel Preparation

2.3.

Histology

The sliced tissue samples were stained with a Masson Trichrome (MT) stain kit (ChromaView advanced testing, 87019, Richard-Allan Scientific, Thermo Fisher Scientific, Wathan, Massachusetts) and imaged (IX51, Olympus).

2.4.

Image Acquisition and Signal-to-Noise Ratio Analysis

Conventional SHG images were taken without pulse compression, in which the laser bypasses the pulse compressor prisms and is routed directly to the microscope. PM-SHIM images were taken when the pulse duration was minimized with optimized pulse compression. All samples, including muscle, liver, and collagen hydrogels, were imaged with both conventional SHIM and PM-SHIM. The average laser power outside the laser was $1.42\mathrm{W}$ and attenuated to $50\mathrm{mW}$ at the sample stage. The PMT voltage was set at $900\mathrm{V}$ for all recordings. In all cases, we recorded a background image using plain glasses as dark background levels for signal processing. We used the $20\times$ objective to obtain $460\times 460\text{-}\mu \mathrm{m}$ , $512\times 512\text{pixel}$ images. Nine such images were stitched to obtain the tile scan image of $1382\times 1382\text{-}\mu \mathrm{m}$ , $1536\times 1536\text{pixel}$ images. The tile scan images were used for analysis purposes. SNR was defined as the average pixel intensity value of SHG signal to the background intensity acquired earlier from the plain glass. SNR of conventional SHIM and PM-SHIM was compared directly by dividing one SNR with the other. The SNR calculation was repeated 50 times, and an average SNR and standard deviation was calculated.

2.5.

Image Acquisition and Quantification of Collagen Remodeling in Tumor Samples

PM-SHIM and conventional SHIM images of the tumor sample were acquired using a $20\times$ objective. Nine images of $512\times 512\text{pixels}$ , $460\times 460\mu \mathrm{m}$ were taken per tumor sample in the tumor interior $200\mu \mathrm{m}$ from the tumor boundary to avoid the collagen concentration spikes in the tumor boundary. Images were acquired at a depth of $20\mu \mathrm{m}$ in the $40\text{-}\mu \mathrm{m}$ tissue section. The tissue section imaged was sliced about $1\mathrm{mm}$ deep into the tumor. An image segmentation algorithm based on a mixture Gaussian model was performed to remove background and noise. It is assumed that the intensity of pixels in the image can be modeled as the mixture of two Gaussian distributions, one representing the collagen area with strong SHG signals, and the other representing the background. Using the expectation-maximization (EM) algorithm,³⁹ the parameters of the Gaussian distributions that model the peak intensity of pixels in the image was found. A binary image was generated by applying a value of 1 to all pixels having intensity that belongs to the Gaussian distribution representing the collagen area, and a value of 0 to the rest of the pixels.

We quantified four parameters, namely collagen area percentage, fiber number, fiber length, and fiber width. The percentage of collagen area was determined as the number of pixels that are segmented as collagen divided by the total number of pixels in the same image. After collagen segmentation, the distance transform is first performed on the binary image, which calculates the distance from a fiber pixel to a background pixel. After thresholding of the distance function, global maximum points are identified as the cross-link points. Starting from these cross-link points, fibers are traced through several local maxima points until the end of the fiber or another cross-link point is reached. The skeleton of each fiber in the image is extracted by connecting cross-link points and local maxima points. Then, the number of fibers, average fiber length, and average fiber width are quantified.⁴⁰ All image processing and algorithm execution were carried out using MATLAB (The Math Works, Incorporated, Natick, Massachusetts). The image processing algorithm code is available for readers on request.

3.1.

Pulse Compressor Optimization

In Fig. 2 , the optimization of prism positions in the pulse compressor to obtain the best pulse modulation is shown. Figure 2 shows the pulse duration plots and Fig. 2 shows the pulse width plots at various prism positions. The optimized prism 1 position was found to be 1100 and the optimized prism 2 position was 2100. The spectral and temporal profiles of the beam measured at three different locations are shown in Fig. 2, in which the beam profile in the PM-SHIM and conventional SHIM are shown in dotted and solid lines, respectively. The pulse width reduced from $14\text{to}13.18\mathrm{nm}$ in the PM-SHIM, and the pulse duration improved from $215\mathrm{fs}$ (conventional SHIM) to $96\mathrm{fs}$ (PM-SHIM). We improved the peak power delivered to the sample from $2.3\text{to}6.5\mathrm{KW}$ after pulse modulation. We can also see that dispersion introduced by the optical components affects mostly the pulse duration instead of the pulse width. The pulse duration measurement at the sample stage was taken from the reflection of the signal from the sample stage. In effect, the light is traveling through the microscope components twice, and the pulse duration measured at the end point was $115\mathrm{fs}$ . Thus, the pulse duration of the beam reaching the sample is estimated to be less than $115\mathrm{fs}$ and more than $96\mathrm{fs}$ (pulse duration measured before entering the microscope).

Fig. 2

Chirp analyses of the laser beam of the PM-SHIM for optimization of prism positions in the pulse compressor. (a) Pulse duration in femtosecond (fs) and (b) pulse width in nanometers for different prism positions with pulse modulation (boxes) and without pulse modulation (triangles) are shown. The maximum pulse duration after dispersion is $215.48\mathrm{fs}$ , and the minimum pulse duration after modulation is $96.31\mathrm{fs}$ . The optimized prism positions are 1100 for prism 1 and 2100 for prism 2. (c) Spatial profile—pulse width in nanometers (left column), and temporal profile—pulse duration in fs (right column) of the laser beam before AOM, after AOM, and at the sample stage are shown for PM-SHIM (dotted line) and conventional SHIM (solid line). The dispersion affects the pulse duration rather than the pulse width.

3.2.

Signal-to-Noise Ratio Improvement in the Pulse-Modulated Second Harmonic Imaging Microscope

SHG images from collagen gels, liver, and muscle sections in the conventional SHIM and PM-SHIM are shown in Fig. 3 . The SHG image obtained from the collagen fibers in the gel construct using the conventional SHIM [Fig. 3] is not clear, while those obtained using PM-SHIM are brighter and sharper [Fig. 3]. Figures 3 and 3 show the portal triads as well as the liver parenchyma in liver lobules imaged with conventional SHIM and PM-SHIM. The SHG signal generated by collagen is shown in the green channel, and the two-photon excited fluorescence (TPEF) in the hepatocytes is shown in the red channel. The smaller collagen fibers in the liver parenchyma are clearly visualized with PM-SHIM but not with SHIM. Similarly, with mouse thigh muscle, the individual muscle fibers and the collagen fibrils surrounding the muscle fibers cannot be visualized with conventional SHIM [Fig. 3], but only with PM-SHIM [Fig. 3].The ratio of SNR from PM-SHIM and SHIM is shown in the lower right corners of the images. On average, there is a $3.3\pm 0.9$ -fold improvement for collagen gels, $2.5\pm 0.7$ -fold increase for liver tissue, and $2.1\pm 0.4$ -fold SNR improvement in muscle samples.

Fig. 3

Collagen gels, liver, and muscle samples exhibit improved SNR with PM-SHIM. Samples demonstrating the improvement of SHG and TPEF signals in collagen gels [(a) and (d)], liver tissue slice [(b) and (e)], and mouse thigh muscle [(c) and (f)]. The degree of SNR improvement is indicated on the improved images in the right bottom corner. The visualization of small collagen fibers in the liver parenchyma (e) and the small collagen fibrils in the muscle sample (f) is made possible with pulse modulation. Scale bar: $50\mu \mathrm{m}$ .

We have demonstrated a marked improvement for collagen visualization with more than two-fold improvement in SNR. As SHG is a stain-free imaging system, the SHG signal intensity observed correlates directly to the collagen amount present in the sample rather than the quantity of dye present in the sample. It also helps in rapid sample preparation, making it an easy technique for imaging biopsy samples, where the tissue can be imaged using PM-SHIM and then used for other routine histology techniques.

3.3.

Collagen Modulation on Drug Administration Visualized with Pulse-Modulated Second Harmonic Imaging Microscope

The in-vivo activity of ABT-869 on MV4-11 xenograft tumors was evaluated previously.¹ The tumors were reduced to unpalpable size but the tumor cells were not completely eliminated by the drug treatment. In the PM-SHIM, the collagen fiber distribution in the drug-treated and control group was clearly visualized [Figs. 4 and 4 ], while in the conventional SHIM very few collagen fibers can be visualized in the tumor stroma, even with maximized laser power and detector sensitivity [Figs. 4 and 4]. The Masson’s Trichrome stain reveals some differences in collagen distribution between treated and control samples [Figs. 4 and 4] in semiquantitative manners.^{41, 42}

Fig. 4

Collagen fibers in chemotherapy drug-treated samples can be clearly visualized using the PM-SHIM. Representative images of tumor samples before and after chemotherapy are shown. (a) and (b) show images taken with conventional SHIM, (c) and (d) show images taken with PM-SHIM, and (e) and (f) show the Masson’s trichrome stained slides of the drug-treated and control samples, respectively. The laser power used to excite the samples with conventional SHIM was 10% higher than with PM-SHIM. The smaller fibers are visualized by the PM-SHIM, which are not excited with conventional SHIM. Scale bar: $50\mu \mathrm{m}$ .

Collagen fiber contents in the tumors were quantitatively analyzed comparing nine images ( $460\times 460\mu \mathrm{m}$ each) from seven treated and seven control tumors. The collagen fiber number and collagen area percentage of the treated and control tumors are shown in Fig. 5 . The drug-treated group is shown by the white bar and the control group by the black bar for both PM-SHIM and conventional SHIM systems. We have found that by using PM-SHIM, the number of collagen fibers is much higher in the drug-treated group [Fig. 5, $3470.8\pm 1092\text{fibers}∕{\mathrm{mm}}^2$ ] than the control group [Fig. 5, $1131.7\pm 315\text{fibers}∕{\mathrm{mm}}^2$ ] with $p<0.0002$ . In the conventional SHIM, the fiber numbers were $1208.6\pm 107.3\text{fibers}∕{\mathrm{mm}}^2$ for the treated group and $386.9\pm 104\text{fibers}∕{\mathrm{mm}}^2$ for the control group with $p<0.079$ . As shown in Fig. 5, by using PM-SHIM the collagen area percentage of the drug-treated group was $7.9\pm 3\mathrm{\%}$ , while that of the control group was $2.0\pm 0.2\mathrm{\%}$ . In the conventional SHIM, the percentages were $2.1\pm 0.7$ and $0.8\pm 0.3\mathrm{\%}$ for the treated and control groups, respectively. On comparing the treated and control samples using a Student’s t-test, the percentage calculated from the PM-SHIM images showed a statistical significance of $p<0.0004$ , and for the conventional SHIM it was $p<0.094$ .

Fig. 5

Quantification of collagen properties in drug-treated samples shows improved fiber number and collagen area percentage with PM-SHIM. The collagen fiber number and area percentage are shown in (a) and (b). The two parameters are plotted for the drug-treated (white bars) and control (black bars) samples with PM-SHIM and with conventional SHIM. Statistical significance was tested with a Student’s t-test, $p<{0.0002}^{**}$ for collagen fiber number and $p<{0.0004}^{**}$ for collagen area percentage with PM-SHIM, and $p<{0.079}^{*}$ and $p<{0.094}^{*}$ , respectively, for conventional SHIM. The difference in the collagen parameters between the drug-treated and control samples can be clearly visualized using the PM-SHIM. The formula used to measure the tumor volume is shown in (c). The tumor volume changes before and after ABT-869 treatment are shown in (d). The percentage volume change per day between the data points are marked in the graph.

On quantifying the collagen fiber lengths and widths, we found that we were able to detect longer and wider fibers in the treated group using PM-SHIM. The longest fiber we detected in the treated group using PM-SHIM was $155.2\mu \mathrm{m}$ , while that of the fibers visualized using conventional SHIM was $48.8\mu \mathrm{m}$ . The longest fiber for the control group visualized using PM-SHIM was $55.7\mu \mathrm{m}$ , and that using conventional SHIM was $32.9\mu \mathrm{m}$ . Similarly, the widest fiber we detected in the treated group using PM-SHIM was $77\mu \mathrm{m}$ , while that of the fibers visualized using conventional SHIM was only $17\mu \mathrm{m}$ . The longest fiber for the control group visualized using PM-SHIM was $12.6\mu \mathrm{m}$ , and that using conventional SHIM was $11.7\mu \mathrm{m}$ . This shows that there were several disconnects in the fibers visualized using conventional SHIM, hence segmenting the same fiber into smaller, thinner fibers. Even though the fibers were segmented, the overall number of fibers detected was not elevated in the conventional SHIM, as many of the fiber signals were too weak to be detected.

The frequency distributions of the length and width of the fibers for the treated group and control group imaged with PM-SHIM and conventional SHIM (solid squares and triangles) are shown in graphs 1 and 2 on the left side of Fig. 6 , respectively. Regions of the plot are enlarged to show clearly the length and width distribution of the fibers in Figs. 6, 6, 6, 6. It is evident from the enlarged graphs that only the length and width distribution of PM-SHIM is distinguishable between the treated and control groups.

Fig. 6

Quantification of collagen fiber length and width shows distinction between the treated and control samples with PM-SHIM (solid squares). Graphs 1 and 2 depict the length and width frequency distribution of the treated and control group visualized using the PM-SHIM (solid squares and triangles) and the conventional SHIM (empty squares and triangles). The enlarged view of collagen fiber length distribution is shown in (a) and (b) and the collagen fiber width in (c) and (d).

In our tumor volume measurements [method to calculate tumor volume is shown in Fig. 5], we find that the percent tumor volume reduction is $24.7\mathrm{\%}\text{per}\text{day}$ in the first two days of treatment, while it decreases to $10.3\mathrm{\%}\text{per}\text{day}$ in the next three days of treatment, as shown in Fig. 5. This drop in percent of tumor volume reduction per day could indicate reduction of efficacy of chemotherapy due to hindrance caused by the wound-healing response. This correlates with our data that collagen increases after treatment.

The fibers observed in the control group include a few thick and long fibers, representing more mature fibers, and some scattered small and thin fibers that appear as speckles, representing less mature fibers or degrading fibers. In the drug-treated group, we observe branched and shorter fibers connected to the long mature fibers, and the fibers appeared to be brighter in general. Not many speckled collagen structures were observed in the drug-treated group. The reduced speckle content in the drug-treated sample could indicate lower degradation of the collagen fiber. The brighter short fibers could indicate more collagen production, and the long thick fibers represent fiber maturation in the drug-treated group. ABT-869 is a multitargeted receptor tyrosine kinase inhibitor targeting mainly the vascular endothelial growth factor receptors (VEGFR) and platelet derived growth factor receptors (PDGFR). PDGFRs have been shown to activate collagen production in sclerosis models.⁴³ Thus, ABT-869 blocking PDGFRs should theoretically down-regulate collagen production, which cannot explain the observed up-regulation of collagen production in the tumors. It is more likely that the chemotherapy triggers a wound-healing response, resulting in the production of new collagen fibers and reduced degradation of the existing fibers.

One of the hallmarks of tumor progression is reduced expression of extracellular matrix, especially collagen type 1.⁴⁴ The collagen in the tumor interior is reduced, while at the tumor boundary the area of collagen (collagen cap) is increased. The increase in collagen in the cap has been attributed to the pushing of the pre-existing collagen bundles by the cancer cells onto the surrounding normal tissue.⁴⁵ The collagen cap is further bolstered by collagen production by activated fibroblasts. The collagen cap acts as a barrier to drugs,²⁴ but when needed the barrier is broken down by metastasizing cancer cells.⁴⁶ In the tumor interior, the collagen fibers are thin and sparse, as they are all newly synthesized by activated fibroblasts and cancer cells but are not part of existing collagen bundles.

During chemotherapy, when the drug diffuses beyond the collagen cap and reaches the tumor interior, several genes in the fibroblasts and cancer cells can be activated to release factors that might render the cancer cells resistant to chemotherapy. There are studies pointing to this effect of chemotherapy on activated stromal cells releasing factors, such as hyaluronic acid,⁴⁷ integrins, and fibronectins,^{48, 49} that are often associated with local wound-healing processes. A study by Farmer ⁵⁰ showed that a distinct increase in the expression of stromal signature genes predicts resistance to chemotherapy in biopsy samples. However, none of these hypotheses have been directly tested by investigating the tumor responses to chemotherapy. In our study, we have quantified an increase in collagen fibers in the tumor interior after chemotherapy, which might be due to the activated stromal cells involved in local wound healing.

Furthermore, this increased collagen in the tumor interior can activate TGF— $\beta$ , a master cytokine which in turn affects the fibroblast growth factor (FGF), platelet derived growth factor (PDGF),⁵¹ insulin-like growth factor (IGF), and Interleukin-6.⁵² These factors exert compounding effects on the proliferation, activation, and transformation of stromal and cancer cells. The collagen increase in the tumor interior can also increase the mechanical stiffness of the tissue microenvironment, which favors cancer cell proliferation.⁵³ Finally, the additional collagen fibers can bind to proangiogenic factors, preventing new vessel formation,⁵⁴ and thus further limiting the access of chemotherapeutic agents to the remaining cancer cells. Therefore, the observed increase in collagen in the tumor interior could impede sustained efficacy of chemotherapy through more complex mechanisms than previously postulated, based purely on the ECM modulation observed in the tumor boundary.²⁴ PM-SHIM provides us with a quantitative tool to further investigate these mechanisms. PM-SHIM can also enable us to design new regimens of drug treatment, including collagen-modulating components introduced at the appropriate time to reduce collagen hindrance and promote drug penetration.

We have demonstrated that the collagen content in the tumor interior is distinctly different after chemotherapy. The mature fibers in the tumor interior can be visualized with the conventional SHIM, albeit with reduced signal intensity, but the small immature fibers that contribute to a considerable amount of collagen area are only visible in the PM-SHIM images. Thus using the PM-SHIM, we obtained accurate quantification of collagen area percentage, fiber number, and collagen fiber length and width, allowing us to draw statistically significant conclusions about the drug effects on tumors.