Tumor Elastography and Its Association with Collagen and the Tumor Microenvironment

Cancer development typically starts with cells acquiring mutations enabling their uncontrolled proliferation and overcoming intrinsic mechanisms regulating tissue homeostasis. This cell-centered view has been expanded over time to include the tumor microenvironment consisting of different stromal cells, immune cells, and the extracellular matrix (ECM) in which all cells are embedded (1–3). The ECM is comprised of a fibrillary protein network (collagen, elastin), glycoproteins (fibronectin), bound growth factors (VEGF, FGF), and glycosaminoglycans [hyaluronic acid (HA)]. Its major roles include the provision of mechanical stability and functional tissue stratification (4, 5).

Although increased stiffness has been recognized as a biomarker for breast cancer associated with a poor prognosis (6–8), the role of tissue stiffness in tumor initiation (9), tumor invasiveness (10), modulation of cell signaling (11), drug delivery (12), and immune cell infiltration has only recently emerged (13–15). These discoveries have fostered experimental approaches that modify the tumor microenvironment, such as preventing collagen crosslinking by blocking lysyl oxidase to reduce tumor stiffness (11), enzymatic degradation of HA to reduce interstitial pressure and improve drug delivery (12) or genetic ablation of carcinoma associated fibroblasts to alter fibrosis and immune cell infiltration (16). The evaluation of therapies targeted at the tumor microenvironment would benefit from noninvasive biomarkers that are associated with therapeutic efficacy. This is particularly important for drugs targeting the ECM, because their effects cannot be measured noninvasively with currently available techniques (5, 17, 18). Tumor stiffness may function as an integrative biomarker of the tumor microenvironment as it is influenced by many factors, including the stiffness of cancer cells, the fraction of stroma cells in tumors, the amount and state of the ECM, the presence of necrotic areas, and the activity of fibroblasts and immune cells.

To enable noninvasive visualization and quantification of stiffness, magnetic resonance and ultrasound-based elastography have been developed (19, 20). Both techniques have been validated clinically for liver fibrosis and breast cancer staging (6, 21–23). However, a systematic evaluation of tumor stromal components and their contribution to stiffness in either clinical or preclinical tumors is currently lacking. A thorough evaluation of the relationship between tumor stromal components and elastography stiffness estimates is needed to robustly employ stiffness as a noninvasive biomarker for therapies that target stromal components. Only a limited number of preclinical tumor elastography studies have been performed so far (24–29). Furthermore, ultrasound-based estimates of tumor stiffness have not been directly correlated with histologic estimates of ECM proteins or collagen content and crosslinking. Because tumor stiffness is an integrative marker of many processes affecting the tumor microenvironment, careful validation is required to map changes in tumor stiffness to its underlying biology. We therefore tracked changes in tumor stiffness and perfusion over several weeks in four syngeneic (B16F10, EMT6, 4T1, KPR3070) and two genetic (Braf/PTEN, MMTV-Her2) tumor models. In vivo stiffness measurements were compared with ex vivo mechanical testing, mass spectroscopy–based estimates of collagen content, and crosslinking. Furthermore, immunofluorescence of tumor microenvironment markers was performed and confocal microscopy images were spatially registered to in vivo US data for analysis of spatial correspondence between these metrics. Finally, localized delivery of collagenase was performed to evaluate ultrasound elastography detection sensitivity for ECM modifications.

An expanded methods section is available in the Supplementary Material.

Tumor cell lines were selected to cover a broad range of tumor stiffness: B16F10 mouse melanoma (ATCC; CRL-6475), EMT6 mouse mammary carcinoma (ATCC; CRL2755), 4T1 mouse mammary gland carcinoma (ATCC; CRL-2539), and in-house mouse pancreatic ductal adenocarcinoma line KPR3070, derived from Kras.lsl.G12D/wt;Trp53lsl/R270H/wt;Pdx.Cre1 genetically engineered mouse model.

All animal procedures were performed according to guidelines from the Institutional Animal Care and Use Committee at Genentech, Inc. Female BALB/c or C57BL/6 (n = 80, n = 45, 20–25 g; The Jackson Laboratory) were inoculated subcutaneously on the right flank with tumor cells (1 × 10⁵ cells/mouse) resuspended in 100 μL Hank's balanced salt solution/Matrigel (BD Biosciences). Additional BALB/c mice were implanted with EMT6 cells + Matrigel (n = 10) or 4T1 cells with or without Matrigel (n = 15, n = 15) in the 5th mammary fat pad (1 × 10⁵ cells/mouse). US imaging was performed 7, 14, and 21 days after inoculation. Animals with tumor ulcerations were euthanized and excluded from the study (see Supplementary Table S1).

BrafLSL.V600E;PTENfl/fl;Tyr.CreER;Rosa26LSL.tdTomato mice were fully backcrossed (>10 generations) into C57BL/6J mice. Licenses were obtained from M. McMahon (UCSF, San Francisco, CA; ref. 30) and H. Wu (UCLA, Los Angeles, CA). At 8 to 12 weeks of age, mice (n = 5) were first anesthetized using 2% isoflurane (Henry Schein). The dorsal skin of the right flank was shaved and 1 μL of 5 mmol/L 4-OH tamoxifen (Sigma) dissolved in ethanol was applied to the shaved skin to induce tumor formation. After application, animals were kept under anesthesia for 3 minutes to allow for ethanol evaporation. Tumors appeared within 4 to 8 weeks, after which weekly ultrasound imaging was performed.

MMTV.huHER2 transgenic animals have been generated in-house and were maintained on a FVB/N background as described previously (31). Following the emergence of palpable tumors (n = 5), weekly ultrasound examinations were performed.

Mice were anesthetized with 4% sevoflurane (Zoetis), lateral tail veins were cannulated and mice were positioned on their left side for syngeneic tumors, prone for Braf/PTEN, or supine for MMTV-Her2 and EMT6 or 4T1 tumors in the mammary fat pad. Following hair removal, anatomic b-mode images (Siemens Acuson S2000, Siemens Medical Solutions) were acquired for axial and sagittal planes covering the maximum tumor cross sections (14L5SP probe, center frequency 14 MHz, 27% power; 50 μm in plane resolution, 300 μm slice thickness, FOV: 3 × 2 cm²). To estimate tumor stiffness, acoustic radiation force impulse (ARFI) imaging was performed on axial planes using parameters optimized for preclinical tumors. Following ARFI imaging, a 9L4 probe (center frequency 8 MHz, 27% power; 90 μm in plane resolution, 300 μm slice thickness, FOV: 3.5 × 3 cm²) was positioned to match the 14L5SP axial plane. Contrast-enhanced ultrasound (CEUS) imaging was performed following infusion of 40 μL of a microbubble contrast agent (SIMB4-5, Advanced Microbubbles Laboratories, CO, 2.3 ± 0.2 × 10⁹ microbubbles/mL) mixed with 60 μL of heparinized saline. CEUS images were acquired for one minute (30 frames/s) starting prior to microbubble infusion. The entire imaging procedure required approximately 15 minutes per animal.

Tumors were excised, core punches (6 or 8 mm diameter, depending on tumor size) were taken, and their height measured using a caliper. A universal testing machine (5943, Instron) was used for unconfined compression of tumor cores. Elastic moduli were calculated for 30% height compression with a linear model fitted to acquired stress–strain curves.

Tumor tissue was reduced with sodium borohydride, hydrolyzed in hydrochloric acid, reconstituted in hexafluorobutyric acid, and subjected to targeted LC-MS/MS on a 6500+ QTRAP (Sciex). Sample analysis was performed in multiple reaction monitoring mode selecting for specific transitions representing hydroxyproline (HYP), pyridinoline (PYD), deoxypyridinoline (DPD), and dihydroxy lysinonorleucine (DHLNL).

Three weeks after 4T1 inoculation, baseline ultrasound (ARFI + CEUS) was performed followed by injection of 20 μL treatment solutions into the core of tumors and at three subcutaneous locations surrounding the tumor. The control group (n = 12) received M2 media (Sigma), while the collagenase group received 0.25 U/mL collagenase-D (Roche). Follow-up imaging was performed 24 hours after baseline imaging and localized delivery.

Following immunofluorescence staining, confocal microscopy was performed using a Nikon A1 microscope (Nikon). Stitched images covering the entire tumor cross-sections were acquired using a 20× air objective (0.41 × 0.41 μm², in plane resolution).

Image analysis was performed using custom, in-house software (Matlab R2016a, MathWorks). Tumor boundaries were manually delineated on axial and sagittal b-mode images to estimate tumor volumes by fitting ellipsoids. Four shear wave speed maps were averaged and converted to elastic moduli: |E\ = \ 3\rho {v^2}$|⁠, where material density |\rho $| (1,020 kg/m³) was assumed. A 5% CEUS enhancement threshold was used to define perfused and nonperfused tumor areas. ARFI artifacts in and around fluid filled cysts in MMTV-Her2 tumors were excluded (Supplementary Fig. S1A).

Images were filtered to reduce noise and thresholds were applied to define stained pixels. Connected objects above a minimum size were used to calculate the percentage of positively stained pixels (collagen, platelet-derived growth factor receptor alpha, and HA). For vessel detection (endomucin/lectin), thresholds were followed by watershed segmentation to separate touching vessels (Supplementary Figs. S2 and S3). Vessel maps were dilated and overlaid on images of alpha smooth muscle actin (SMA) stained tissue to define vessel-associated SMA and nonvessel SMA considered activated fibroblasts/myofibroblasts (Supplementary Fig. S3). To estimate T-cell (CD3) and macrophage density (F4/80), thresholds were applied and objects within single-cell size range were classified as cells (Supplementary Figs. S2F and S4).

Manual regions of interest were drawn on parametric confocal maps to remove nontumor tissue (skin, fat) and maps were registered (translation, rotation and scale, Supplementary Fig. S5) to ultrasound data.

For multiple regression analysis, tumor masks were applied to all ultrasound and confocal microscopy–derived parametric maps to generate animal-specific averages. Pixel-wise correlations were performed by combining parameter vectors for all animals with the same tumor type. To facilitate the comparison of microenvironment parameters within and across different tumor types, 2D parametric maps were reduced to 1D line representations based on the relative distance between center and rim of individual pixels (Supplementary Fig. S6).

Results are shown as mean ± SEM. For the comparisons of different tumor types, two-tailed Wilcoxon rank sum tests were used. To test for linear relations between different parameters, multiple or simple linear regression analysis was performed. No regression lines were plotted for P > 0.05 or r < 0.2. To test for differences in tumor volume, perfusion, or stiffness with time, one-way ANOVA was performed or two-way ANOVA for temporal differences between tumor types was applied. Grubbs outlier tests were performed on ARFI stiffness and hydroxyproline measurements for the combined mass spectroscopy dataset. Based on P < 0.01, six measurements were removed from the dataset (n = 107). Correlation maps were generated using the R corrgram package (v1.12) showing all correlations with P < 0.1. Statistical analysis was performed using R software version 3.2.4. The following significance levels were used: *, P < 0.05; **, P < 0.01; *** P < 0.001.

To evaluate the consistency of ARFI stiffness measurements in preclinical tumor models, tumor imaging was repeated within 2 hours. The coefficient of variation for tumor stiffness measurements was 12.5% (Supplementary Fig. S7A, B). Syngeneic tumors frequently develop nonperfused/necrotic regions in the center (core). To account for this, perfusion maps were used to define perfused and nonperfused tumor tissue on stiffness maps (Fig. 1A–D). Although soft (B16F10) tumors presented with increased stiffness around the tumor rim after the first imaging time point, stiff tumors (4T1, KPR3070) showed increased stiffness throughout the tumor including nonperfused cores. (Supplementary Figs. S7 and S8). Tumor growth was fastest for B16F10 melanomas and slowest for KPR3070 tumors (Fig. 1E). As tumors grew, the fraction of perfused tissue declined (Fig. 1F). In perfused tumor tissue, stiffness increased with time for slow-growing KPR3070 and 4T1 tumors. In contrast, fast growing B16F10 showed no stiffness increase (perfused area).

Tumor volumes for EMT6 and 4T1 tumors were similar after growing for 2 weeks in the mammary fat pad or 3 weeks in the flank (Fig. 1I), reflecting faster growth at their orthotopic site. There was no difference in perfused tumor stiffness at these time points for 4T1 tumors. However, EMT6 tumors grown orthotopically had a higher stiffness (Fig. 1J) despite faster growth at their orthotopic site. To assess the potential impact of Matrigel on tumor stiffness and perfusion, 4T1 tumor cells were implanted orthotopically with or without Matrigel. Although Matrigel had no significant effect on tumor growth, tumor perfusion was significantly lower for 4T1 tumors implanted with Matrigel (Supplementary Fig. S9A–S9C). Furthermore, there was no significant difference in tumor stiffness for perfused or nonperfused areas of 4T1 tumors implanted with or without Matrigel in the 5th mammary fat pad (Supplementary Fig. S9D and S9E).

Tumor models could be clearly separated based on ultrasound stiffness measurements (average of perfused and non-perfused) at the final imaging time point (Fig. 2A), leading to a ranking of increasing stiffness from B16F10 to EMT6 to 4T1 and KPR3070. Following in vivo imaging, ex vivo mechanical testing was performed. Ranking tumors based on ex vivo mechanical stiffness measurements matched the ultrasound-based ranking, but ex vivo differences between EMT6 and 4T1 were not statistically significant and KPR3070 tumors were substantially stiffer than 4T1 tumors (Fig. 2B). Mechanical testing of KPR3070 may have led to higher measurement errors due to small tumor volumes (see Discussion). Although regression analysis found a good correlation between mechanical testing and US stiffness, this correlation could be significantly improved by excluding KPR3070 tumors (Fig. 2C and D).

Mass spectroscopy of excised tumors revealed a strong correlation between HYP content (a marker for collagen content) and in vivo ultrasound stiffness measurements (average of perfused and nonperfused; Fig. 2E). HYP content increased from soft B16F10 to stiff KPR3070 tumors matching ultrasound-based stiffness expectations, except for EMT6, which had a higher than expected HYP content (Fig. 2F). Although mature PYD and deoxypyridinoline (DPD) crosslinks were correlated with tumor stiffness when normalized to tumor mass (Supplementary Fig. S10), these correlations disappeared when normalized to collagen content (Fig. 2G). Furthermore, PYD and DPD crosslinks normalized to collagen were not different between tumor models. In contrast to mature crosslinks, immature DHLNL crosslinks were correlated with tumor stiffness even when normalized to collagen content (Fig. 2H). Accordingly, including HYP and DHLNL into a linear regression model for ARFI stiffness increased the multiple correlation coefficient to 0.68 (P < 0.001).

Unlike syngeneic models, Braf/PTEN melanoma and MMTV-Her2 breast cancer were well perfused and did not develop necrotic areas (Fig. 3A–D; Supplementary Fig. S1). However, MMTV-Her2 tumors developed progressively, which included enlargement of fluid filled cysts. These cysts were excluded from stiffness measurements if they were detectable on b-mode ultrasound images (Supplementary Fig. S1A). Tumor volumes increased linearly with similar rates in both models (Fig. 3E). Although Braf/PTEN tumors were almost completely perfused at the two time points assessed, perfused tumor areas decreased slightly with time in MMTV-Her2 tumors (Fig. 3F). Nonetheless, this decrease may be due to developing cysts, which are not detectable on b-mode images rather than decreased perfusion in viable tumor tissue. In contrast to Braf/PTEN melanoma, where stiffness in perfused tumor tissue did not change with time, a small increase was observed for MMTV-Her2 tumors (Fig. 3G). Similarly, stiffness in the nonperfused tissue of MMTV-Her2 tumors increased with time albeit not significantly (Fig. 3H).

Immunofluorescence showed a large number of apoptotic cells in syngeneic tumors, but only few in genetic tumors (Fig. 4). These were typically found in the center of EMT6, 4T1, and KPR3070 tumors in line with a low density of perfused vessels (lectin). In contrast, B16F10 tumors contained pockets of apoptotic cells surrounded by large perfused vessels lacking capillaries throughout the tumor cross-sections. Although collagen (collagen I + III + IV) in soft tumors (B16F10, Braf/PTEN) was primarily found in vessel basement membranes, stiff tumors (4T1, KPR3070) showed increasing amounts of collagen throughout the tumor cross-sections including apoptotic regions in the center. Similarly, soft tumors contained only vascular smooth muscle cells (αSMA⁺), whereas stiff tumors were characterized by abundant myofibroblasts (nonvascular αSMA⁺) in the perfused tumor areas. Classical fibroblasts (PDGFR-α⁺) were primarily localized in the tumor rim often associated with high densities of T cells (CD3⁺).

To facilitate a more systematic analysis of ultrasound-based perfusion and stiffness measurements and their biological relevance, automatic image segmentation and registration of confocal images to ultrasound data was performed (Supplementary Fig. S5). As a first analysis level, cross-section averages for different parameters from all tumor models used in this study were combined to generate a correlation matrix (Fig. 5A). Stiffness was positively correlated with collagen, vessel, myofibroblast, and fibroblast density and negatively correlated with tumor volume, relative blood flow, and viability. Collagen density and tumor volume had the strongest positive and negative correlations with stiffness, respectively (Fig. 5B–E). Furthermore, stiffness was also positively correlated with T-cell density, but no correlation was found between stiffness and cell density (Supplementary Figs. S11 and S12). Multiple regression analysis showed that a combination of collagen density, vessel area density, and relative blood flow could best model tumor stiffness (r = 0.76, P < 0.001). Although comparisons of cross-section averages are informative about the relevance of different parameters, they cannot address their spatial distribution within tumors. We therefore performed pixel-wise correlations between T-cell density and different matrix parameters. In KPR3070 tumors, T-cell density was positively correlated with relative blood flow and macrophage density, but only weak correlations were found between T-cell and fibroblast or collagen density (Fig. 5F–I). EMT6 and 4T1 showed similar T-cell distribution patterns, except that collagen was positively correlated with T-cell density in EMT6 tumors. In contrast to these stiffer tumors, soft B16F10, MMTV-Her2, and Braf/PTEN tumors only showed significant correlations between macrophage and T-cell density (data not shown). To further illustrate the spatial distribution of tumor microenvironment parameters and to facilitate the comparison of different tumor models, 2D images (parametric maps) were reduced to a line representation based on the relative center-rim distance of individual pixels (Supplementary Fig. S6). Line plots illustrate that stiffness in EMT6, 4T1, and KPR3070 tumors was higher in the center while soft B16F10 tumors were stiffer in the tumor rim but soft in the center (Fig. 5J). This center versus rim pattern was reversed for blood flow and vessel density but similar for collagen density (Fig. 5K–M). Myofibroblast, fibroblast, macrophage, and T-cell density qualitatively tracked with blood flow and vessel density (Fig. 5N–Q), indicating the importance of perfused vessels for their distribution.

To assess the sensitivity of ultrasound elastography to perturbations of the ECM, localized injections of collagenase were performed (Supplementary Fig. S13A). Injections were performed after baseline ultrasound imaging with follow-up 24 hours thereafter. Tumor perfusion and tumor volume did not change significantly in treatment or control groups (Fig 6B, Supplementary Fig. S13B and S13C). However, collagenase-treated tumors showed a significant decrease in stiffness (Fig. 6A, C, and D). This decrease in tumor stiffness was corroborated by decreased HYP content in collagenase-treated tumors (Fig. 6E), but there was no difference in collagen crosslinks normalized to tumor weights between groups (Supplementary Fig. S14).

Stiffness can be measured invasively at a cellular, tissue, or organ level. Although cellular measurements have shown that tumor cells are typically softer than normal cells, most tumors present with increased stiffness compared with the surrounding tissue. These changes can be quantified with elastography noninvasively, providing useful information about the tumor microenvironment particularly the ECM and its modification by stromal cells. Ultrasound elastography is the most frequently used technique to assess tumor stiffness due to the ease of use and relatively low cost. When combined with contrast-enhanced ultrasound, perfusion and stiffness can be assessed within one imaging session clinically and preclinically. Magnetic resonance elastography (MRE) offers the advantage of three-dimensional stiffness and viscosity measurements over a wide range of mechanical actuation frequencies with low variability. However, MRE is not as widely available as ultrasound elastography. Other disadvantages of MRE are substantially longer imaging times and higher costs. MRE can be combined with tissue perfusion and vascular permeability quantification within one imaging session. Another technique that has been used preclinically is second harmonic imaging of collagen (32), but this is only a surrogate for stiffness and requires invasive optical windows.

We have shown that ultrasound elastography can reproducibly measure tumor stiffness in syngeneic and genetic tumor models and that in vivo stiffness measurements correlate well with ex vivo mechanical testing. Although reproducibility data for preclinical ultrasound elastography was not previously available, the observed coefficient of variation was similar to estimates from preclinical MRE and clinical US elastography (33, 34). In addition, we found a good correlation between unconstrained mechanical compression and ARFI stiffness measurements. This correlation improved further when KPR3070 tumors were excluded. Mechanical compression required standardization of tumor samples to cylinders of known diameter and height, which were generated by taking core punches from excised tumors. Although diameter-to-height ratios were close to one for most tumors, KPR3070 tumors had ratios close to two due to their size and shape. This limitation is likely responsible for the high mechanical stiffness estimates for KPR3070 tumors. Furthermore, the use of core biopsies to estimate mechanical tumor stiffness introduces sampling errors particularly for soft tumors as most of their mechanical strength is localized in the tumor rim. The ability of elastography techniques to noninvasively measure tumor stiffness in their undisturbed environment should circumvent these problems.

In line with previous publications (25, 26, 35), we found strong correlations between tumor volume and stiffness for the slower growing tumors in this study such as 4T1 and KPR3070. However, tumor volume and stiffness were not correlated in B16F10 or the genetically engineered Braf/PTEN mouse melanoma models. Furthermore, tumor volume and stiffness were only weakly correlated in syngeneic EMT6 and genetically engineered MMTV-Her2 tumor models. These results indicate that stiffness in fast growing syngeneic tumors is likely to reflect cancer cell stiffness, whereas in slow growing syngeneic tumors, stiffness is dominated by ECM and stromal cell changes. Accordingly, tumor models may be grouped into these two general classes and, thus, simplify the interpretation of observed stiffness changes.

The site of tumor implantation may also affect elastic properties of tumors due to local stroma differences. Nonetheless, tumor stiffness for volume-matched 4T1 tumors grown at orthotopic or ectopic sites was not different. In contrast, EMT6 tumors were significantly stiffer when grown orthotopically despite faster growth. This indicates that effects of the implantation site on tumor stiffness are cell line dependent, which may explain the differences observed in the literature (35, 36). Furthermore, different tumor models implanted in the flank led to substantial stiffness differences, highlighting the importance of the tumor line over implantation site. The use of Matrigel as an implantation carrier is an additional variable that may influence tumor stiffness. However, we did not detect any stiffness differences between 4T1 tumors implanted orthotopically with or without Matrigel.

CEUS perfusion estimates were used to identify perfused and nonperfused regions of tumors since these regions may differ in stiffness. Syngeneic tumor models showed a continued decline in perfused tumor area as growth outstripped vascular supply. Stiffness measurements in nonperfused areas were generally higher compared with perfused areas except for day 7, which probably reflects low levels of extracellular matrix deposition in the remaining Matrigel plug. Increased stiffness in nonperfused areas at later time points indicate fibrosis that has not yet progressed to coagulative necrosis, which is associated with matrix destruction and reduced stiffness (25, 26). Perfusion measurements also highlighted the well-perfused nature of genetic compared with syngeneic tumors. Syngeneic B16F10, which despite higher perfusion rates than most syngeneic tumors, did develop nonperfused areas, while Braf/PTEN tumors maintained perfusion throughout the tumor for the duration of the study.

Collagen is the most abundant protein in the ECM, providing most of the mechanical strength in tumors. Collagen cross-linking has been implicated as an important mechanism by which cancer and stromal cells influence the mechanical properties of their environment (11). We found a good correlation between tumor stiffness and HYP, a measure of collagen content in line with published data (11, 16, 37, 38). Furthermore, immature DHLNL collagen crosslinks normalized to collagen content were positively correlated with stiffness, whereas normalized mature DPD and PYD crosslinks were not. The link between collagen accumulation, increased interstitial pressure, and reduced perfusion has been explored in the past (17, 39). However, collagen content and the degree of crosslinking for multiple tumor models have not been compared previously. Tumor stiffness was primarily determined by collagen content and to a smaller degree by immature crosslinks. Interestingly, there was no difference in mature crosslinks normalized to collagen between the tumor models that we analyzed. Inhibition of collagen crosslinking has been shown to reduce tumor growth, metastasis, and invasion (11, 37, 38). Enzymes responsible for collagen crosslinking affect immature crosslinks. It is therefore likely that elastography could track such changes in vivo.

The importance of collagen for tumor stiffness was further corroborated by immunofluorescence and automatic segmentation, which showed strong correlations between collagen density and tumor stiffness on a tumor level and spatially within tumors. In addition, tumor stiffness was correlated with myofibroblast, fibroblast, and vessel density on a tumor level. Positive correlations between microvascular density and tumor stiffness have been reported in the past (27, 35) supporting our findings. These studies also observed positive correlations between cell density and tumor stiffness, which we failed to observe. Such differences may be due to the relatively soft tumor models used for these studies (<7 kPa), which have low amounts of nonvascular-associated ECM increasing the relative importance of cancer cell stiffness for tumor stiffness.

Pixel-wise analysis showed negative correlations between myofibroblast or vessel density and stiffness for stiff tumors such as 4T1 and KPR3070 and positive correlations for soft B16F10 and Braf/PTEN tumors. This pattern may be due to differences in perfusion. Although cores of stiff tumors were nonperfused and apoptotic, soft tumors had perfused cores with higher cell viabilities. Pixel-wise correlations estimated tumor model–specific ultrasound detection sensitivities for perfusion measurements. While approximately 45 perfused blood vessels per square millimeter were required in KPR3070 tumors, 15 were sufficient in B16F10 tumors for CEUS-based detection. One of the limitations of our study is the lack of exogenous fiduciary markers, which may have led to errors during the registration of in vivo ultrasound data to ex vivo histologic data. Pixel-wise correlations between in vivo and ex vivo parameters are particularly affected by such errors. Areas classified as nonperfused based on ultrasound measurements include underperfused hypoxic/necrotic areas (40, 41). Hypoxia leads to fibroblast activation and increased ECM synthesis (42, 43), which may explain the increased stiffness in nonperfused tumor areas.

The relationship between increased tumor stiffness, high interstitial pressure, reduced tumor perfusion, and reduced drug delivery is well documented (17, 39, 44). Limited perfusion of tumor cores may also impede immune cell access and survival (45, 46). More recently, collagen density and crosslinking have been proposed as additional physical barriers impeding T-cell migration (15). Our pixel-wise correlations found strong correlations between perfusion and T-cell density as well as fibroblast and T-cell density. However, collagen density and T-cell density showed only weak negative correlations for KPR3070 and 4T1 tumors and weak positive correlations for EMT6 and B16F10. Furthermore, T-cell density in tumor rims peaked after collagen density started to decline. This indicates that for tumor models analyzed in the current study, blood flow, vessel density, and fibroblast density were important predictors of T-cell density, whereas collagen was not.

Previous studies have shown that MRE and US elastography are able to detect changes in tumor stiffness following cytotoxic or antivascular agent administration in preclinical tumor models (26, 33, 35). However, the ability of elastography to detect direct ECM modifications in vivo has not been documented so far. Our study found that localized collagenase delivery led to a significant reduction in tumor stiffness without affecting tumor volume or perfusion. In agreement with reduced stiffness, mass spectroscopy detected lower collagen concentrations in treated tumors. This indicates that US elastography can detect collagen degradation via reduced mechanical stiffness.

Our findings may help to inform the clinical interpretation of tumor stiffness and stiffness changes following therapeutic interventions, where elastography can provide a biomarker for therapies that target stromal components. However, it will require clinical validation as preclinical tumor models, although informative, do not fully replicate the complexity of human tumors.

In summary, we have shown that ultrasound elastography can reproducibly assess stiffness in syngeneic and genetic preclinical tumor models. Tumor stiffness was strongly correlated with total collagen content and immature collagen crosslinks. Histologic analysis showed that stiffness was correlated with collagen, myofibroblast, fibroblast, and vessel density. Furthermore, T-cell density was primarily determined by vessel and fibroblast density with limited importance for collagen density. We also demonstrated the ability of elastography to detect ECM changes following collagenase delivery. These results show the utility of elastography as a microenvironment biomarker.

J. Riegler is an employee of Genentech. R.A.D. Carano holds ownership interest (including patents) in Roche. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Riegler, S. Rosenzweig, A. Castiglioni, S.J. Turley, J. Schartner, R.A.D. Carano

Development of methodology: J. Riegler, S. Rosenzweig, R.A.D. Carano

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Riegler, V. Javinal, A. Castiglioni, C.X. Dominguez, J.E. Long, Q. Li, W. Sandoval, M.R. Junttila, S.J. Turley, J. Schartner

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Riegler, Q. Li, W. Sandoval, S.J. Turley, J. Schartner, R.A.D. Carano

Writing, review, and/or revision of the manuscript: J. Riegler, Y. Labyed, S. Rosenzweig, A. Castiglioni, J.E. Long, Q. Li, M.R. Junttila, S.J. Turley, J. Schartner, R.A.D. Carano

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.A.D. Carano

Study supervision: J. Schartner, R.A.D. Carano

We would like to thank Franklin Peal for help with interpretation of histologic findings and Cecile Chalouni for her help with confocal microscopy.

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.