Stretching inhibits tumor growth in MMTV-PYMT via a direct mechanical effect.

Background
Mechanical forces such as tension, compression, and shear are generated across many magnitudes and scales in biology, from the cell level to the organismal level, regulating cell fate, guiding developmental processes, and changing cell and extracellular matrix mechanics [16, 37]. In many established models, biomechanical cues impact the growth and invasiveness of cancer cells, with the most consistently demonstrated effects being enhanced tumorigenesis induced by increased extracellular matrix stiffness and mechanical forces [3–6]. In addition, there is mounting evidence that the architecture of the tumor-associated stroma influences the growth and invasiveness of cancer cells [8, 9, 14, 31]. Collagen fibers aligned perpendicular to the boundary of mammary tumors [12] promote stromal invasion by cancer cells in mouse mammary tumor models and are associated with a poor prognosis in human patients with breast cancer [9]. Conversely, collagen fibers oriented parallel to the tumor boundary are found predominantly in more benign neoplastic lesions. Collagen fiber orientation within and around tumors is hypothesized to reflect the balance of outward-directed and inward-directed forces generated by the growing tumor and the surrounding matrix [29]. In response to increasing compressive forces, tumor cells gradually remodel the cancer-associated stroma, creating perpendicular “cables” that facilitate the outward migration of tumor cells and metastasis [31].
In this context, we previously reported a 50% reduction in tumor growth in response to in vivo whole-body stretching (~ 25% strain for 10 min twice a day) in p53/PTEN − / − orthotopic mouse mammary tumors [4]. Considering the large body of evidence showing increased tumor growth in response to increased mechanical forces, this seemingly paradoxical result suggested that further studies of this relatively short duration, moderate amplitude stretching model are warranted.
In the present study, we aimed to further understand the impact of stretching on tumor growth by investigating (1) the effect of stretching in a more invasive mouse model of breast cancer (in vivo and in vitro), (2) the mode of stretching (active and passive) and its orientation relative to the location of the tumor, (3) the dose of stretching required to reduce the size of tumors, (4) whether stretching can modify the tumor-associated extracellular matrix, and (5) the impact of stretching on the tumor cellular microenvironment landscape. We investigated these questions in the well-characterized MMTV-PYMT (mouse mammary tumor virus-polyoma middle tumor-antigen) caused by the expression of the polyoma middle T oncoprotein (PYMT) in the mammary epithelium. MMTV-PYMT mice develop spontaneous mammary tumors that closely resemble the progression and morphology of human breast cancers [34]. In the orthotopic model used in this study, Met-1 cells derived from MMTV-PYMT mammary tumors were injected subcutaneously into the back at the same location as in our previous study [4]. We also tested the effects of stretching in a transgenic MMTV-PYMT model in which tumors developed spontaneously. In the orthotopic and transgenic models, we examined the effects of both active stretching in conscious animals and passive stretching under anesthesia compared with anesthesia alone without stretching. Finally, we investigated the effect of stretching in vitro in MMTV-PYMT tumor spheroids embedded in collagen gels.

Results
In the orthotopic model, Active Stretch for 10 min twice a day resulted in significantly smaller tumor volumes in the Stretch group compared to the No Stretch group 1, 2, and 3 weeks after tumor cell injection, as measured by both caliper (Fig. 1B) and ultrasound (Fig. 1C) (main effect of stretch: p < 0.0001 for both caliper and ultrasound measurements). Three weeks after injection, Active Stretch reduced average tumor volume by ~ 40% and ~ 33% based on caliper (p > 0.001) and ultrasound (p < 0.05) measurements, respectively, compared with the No Stretch control group. As expected, the caliper measurements of tumor volume (including those of the tumor and surrounding tumor-associated stroma) were greater than those obtained by ultrasound (including those of the tumor only), and the measurements obtained via the two methods were positively correlated (r = 0.65).
Because the Active Stretch intervention involves holding the animals by their tails, we wanted to rule out stress as a confounding factor between the Stretch and No Stretch interventions. We therefore conducted experiments in which animals were stretched under anesthesia by passively stretching the whole body longitudinally by ~ 35% of its initial length (Fig. 1D) compared with anesthesia alone without stretching. We observed that Passive Stretch resulted in a ~ 50% reduction in tumor growth compared with No Stretch, starting at week 1 through week 2 following injection, as measured by both caliper (Fig. 1E) and ultrasound (Fig. 1F) (main effect of stretch p = 0.004 and p = 0.02 for caliper and ultrasound, respectively).
To further understand the local impact of mechanical intervention, we used the MMTV-PYMT transgenic mouse model in which tumors develop spontaneously; this is the most used genetically engineered mouse model for cancer research. Mammary lesions arising in MMTV-PyMT mice exhibit molecular and histological progression similar to that of human breast tumors, making them an invaluable tool for cancer researchers and instrumental in understanding tumor biology [22]. We first tested the effect of Active Stretch for 10 min twice a day from 6 to 8 weeks of age, and tumors were measured by ultrasound (as tumors are too small at that stage to be measured by caliper in this model). We observed that Active Stretch did not change the size of the tumors compared to the No Stretch group (main effect of stretch p = 0.32) (Additional file 3A). Furthermore, passive stretching extends the limbs longitudinally and thus mimicking the posture of the animal during active stretching, also did not change the size of the tumors in the transgenic model (Additional file 3B), p = 0.34. We hypothesized that the lack of impact of stretching in the transgenic model may be related to the ventral location of transgenic tumors (Additional file 2), compared with the orthotopic tumors, which are located more dorsally.
Tumors implanted on the dorsal side of the animal lay over the thoracolumbar fascia which connects the shoulders and hips and experiences increased shear strain during longitudinal stretching (active or passive) when the shoulder-to-hip distance is increased. In contrast, tumors located ventrally overlay the pectoral and abdominal fasciae which connect shoulder-to-shoulder and hip-to-hip, respectively, and are thus less mechanically impacted by longitudinal stretching and more by transverse stretching that increases shoulder-to-shoulder and hip-to-hip distance. Therefore, we designed a new Passive Stretch protocol under anesthesia, where the limbs are stretched on the transverse axis (Fig. 2A). Using this procedure, we observed that Passive Stretch reduced the size of the transgenic tumors relative to No Stretch (Fig. 2B) (main effect of stretch, p < 0.0001). This suggests that engaging the tissues in an appropriate direction relative to the location of the tumor is important.
Next, we investigated the effect of varying doses of stretching to reduce the size of the tumors, using longitudinal passive stretching under anesthesia in the orthotopic model. By measuring the distance of longitudinal stretching as the distance between front and rear toes in the prone position, we applied a range of 25–40% stretching amplitudes (strain) compared to No Stretch. A 30% amplitude was maximally effective in reducing the size of the tumors (p < 0.05), while the largest dose (40%) did not reduce the size of the tumors and was comparable in size to the No Stretch group (Fig. 3).
Because of the well-established importance of the extracellular matrix in mechanotransduction, our first step in investigating the mechanism of stretching was to measure the RNA expression of genes linked to fibrosis, mechanosensors, matrix connections, and remodeling: Col1a1, Col5a2, Col6a2, aSMA, TGFb, Piezo1, Connexin 43, DDR1, and DDR2. As shown in Fig. 4, we detected no significant differences in gene expression between the Stretch and No Stretch conditions for any of these markers in either the Active Stretch (Fig. 4A) or Passive Stretch (Fig. 4B) models, except for Col5A2, which was slightly reduced (~ 30% reduction) in the Active Stretch group compared with the No Stretch group.
We then hypothesized that stretching could impact the organization of the extracellular matrix architecture, specifically the tumor-associated collagen matrix signature (TACS), which was previously reported to be involved in the progression of tumorigenicity [5] (Fig. 5). In a subset of animals with orthotopic tumor injections randomized to Active Stretch vs. No Stretch (N = 16), the percentage of TACS 2 collagen fibers oriented parallel (< 15-degree angle) to the tumor boundary were increased (p = 0.02), whereas the number of TACS 3 collagen fibers aligned perpendicular (> 75-degree angle) to the boundary of tumors decreased (p = 0.02) in the Stretch group compared with the No Stretch group (Fig. 5C). These differences resulted in the ratio of TACS 2 to TACS 3 fibers being significantly greater in the Stretch group (p = 0.03) (Fig. 5D), suggesting a more benign tumor phenotype.
Next, we hypothesized that the reduction in the size of the tumors could be related to changes in the tumor immune microenvironment since reorganization of the collagen fibers in the ECM could potentially impact immune cells in the tumor stroma, in addition to tumor cells themselves. We performed spatial transcriptomics on 20 tumor samples (10 No Stretch and 10 Stretch) using a panel of 100 probes to study the tumor cell composition and observed no statistically significant differences between No Stretch and Stretch conditions regarding the composition and distribution of intratumoral immune cell populations, with the exception of a slight increase in epithelial luminal cells in the Stretch group (Fig. 6A, B, Table 1).

Previous studies have demonstrated that tumor spheroids can be used to study mechanical factors influencing tumor cell behavior in vitro [32]. Therefore, we used an in vitro model consisting of only tumor cells and a collagen matrix to further explore the impact of stretching on tumor cell behavior in the absence of other influences that are present in vivo, such as immune, vascular, lymphatic, and nervous systems. Tumor spheroids were created using the Met-1 MMTV-PYMT cell line and then embedded in collagen gels (2.3 mg/ml). Collagen was glycated 6 days prior to beginning the experiment to increase its stiffness, improve cell adhesion to the matrix, and mimic the viscoelastic properties of interstitial connective tissue in vivo. This also allowed the gel to be robust enough to withstand stretching. Collagen-embedded spheroids were randomized to Stretch (23% strain) vs. No Stretch for 10 min twice a day for 3 days to mimic the in vivo procedure (see methods for details). On day 3, we observed a 34% reduction in the spheroid crown area (p = 0.01) (Fig. 7F), a 35% reduction in the number of migrating single cells (p = 0.01) (Fig. 7G), and a 28% reduction in the maximum distance traveled by migrating cells (p = 0.03) in the Stretch gel compared with the No Stretch gel (Fig. 7H).

Discussion
In this study, we showed that daily stretching reduced tumor growth in MMTV-PYMT orthotopic mammary tumors implanted subcutaneously using both Active and Passive Stretch and in MMTV-PYMT transgenic mammary tumors using Passive Stretch. The magnitude of the reduction of tumor volume in response to stretching was between 30 and 50% in the orthotopic (Active and Passive Stretch) and transgenic (Passive Stretch) MMTV-PYMT models compared with the no stretch model. This effect is similar in magnitude to the effect we observed in the P53/PTEN − / − model [4]. We observed a 34% reduction in the spheroid crown area in vitro after only 3 days of stretching, suggesting that stretching can have a very rapid effect on tumor invasiveness.
Several aspects of this study support a mechanical effect of stretching on tumor growth: (1) tumor growth was reduced by Passive Stretch under anesthesia, compared with anesthesia alone, which controls for the effects of stress and muscle activity; (2) tumor localization in relation to the axis of stretch appears important for reducing the size of the tumors, as suggested by the transgenic Passive Stretch model; (3) the more benign tumor-associated collagen signatures (increased TACS2/TACS3 ratio) with stretching suggest a local impact of stretching on organization and mechanical properties of the tumor-associated connective tissue matrix [9]; (4) the reduced spheroid crown area and tumor cell migration following stretching demonstrated effects of stretching in vitro that are directly local and mechanical.
What plausible mechanism may link the application of a mechanical force (stretching) to reduced tumor growth? An important clue to answer this question may be the consistent observation across all of our studies that, in Stretch animals, tumors are easy to dissect away from surrounding tissues and come out resembling “little pearls”. In contrast, in No Stretch animals, tumors are more adherent to the surrounding tissues. This observation may be particularly relevant given (1) the anatomical location of the tumors in our experiments and (2) the shear strain created by stretching within subcutaneous connective tissue. As illustrated in Fig. 8, mammary tumors (both spontaneously occurring and orthotopically implanted in this study) are located within a layer of areolar subcutaneous connective tissue “sandwiched” between two denser connective tissue layers: the subcutaneous fascia (superficial to the tumor) and the deep thoracolumbar (orthotopic model) and pectoral (transgenic model) fasciae. Elongation of the limbs (longitudinal or transverse) pulls on the deep fasciae which are attached to the shoulders. Because of the layered organization of subcutaneous tissues, this pulling motion on the deep tissues creates shear strain relative to the superficial fascia. This in turn results in shear torsion of the tumor relative to the surrounding loose subcutaneous tissue (Additional file 4).
Shear torsion of the tumor relative to surrounding tissues could be an important mechanism disrupting the early attempts of tumor cells to migrate away from the tumor. As tumors grow, individual cancer cells reorganize the matrix and create “tracks” perpendicular to the tumor that facilitate the movement of subsequent cells (Fig. 9B, C) [36]. As illustrated in Fig. 9D, shear strain during stretching may disrupt the initial delicate TACS 3 perpendicular fiber bundles and impair the early cancer cell migration attempts. In stretched animals, instead of migrating, cancer cells may continue to proliferate and exert increased pressure on circumferential TACS 2 fibers (parallel to the tumor) which has been shown to encourage tumor encapsulation [13], hence, the “little pearl” phenotype observed in the Stretch animals. In contrast, tumors in No Stretch animals exhibit perpendicular TACS 3 fibers that attach the tumor to the surrounding matrix and encourage invasion.
The reduction in crown area and single cell migration observed in our spheroid model demonstrated that a beneficial effect of stretching can be demonstrated in vitro. However, in the current in vitro experiment, the applied force and resulting strain were axial, rather than in shear, and therefore were missing an important biomechanical element present in vivo. This limitation could be feasibly addressed in future studies using a three-layered tissue chip (two dense outer layers and tumor cells implanted in a loose matrix middle layer), applying shear strain across the two outer layers and measuring fiber alignment and tumor cell migration within the middle layer.
Our spatial transcriptomics experiment showed no marked differences in the proportion of immune cell populations between no stretch and stretch samples, consistent with the data observed in our previous study, suggesting that immune modulation may not be the primary driver of the response to stretching [4]. Nevertheless, additional experiments aimed at studying the impact of stretching on the more detailed compartment distribution of immune cell populations within the tumor microenvironment are warranted to further explore the potential impact of stretching on immune responses within the tumor.
Overall, the observed reductions in tumor growth following stretching in our experiments may seem surprising, given that most of the cancer mechanotransduction literature shows detrimental effects of mechanical stimulation on cancer [21, 23, 33]. However, there are important aspects of our experiments that may explain this seemingly paradoxical effect. First, in contrast to published studies showing a negative impact of mechanical forces generated by cells within the tumor microenvironment (e.g., myofibroblasts), the mechanical force in our study comes from outside the tumor-associated tissues (i.e., skeletal muscle contraction in the active stretching model and externally applied force in passive stretching model) and may override the pathological matrix remodeling induced by the tumor-associated cells. Second, the duration of stretching used in our experiments was relatively short (10 min) compared with studies in which prolonged cyclical stretching was applied for several hours [2, 26]. Third, the mode of delivery of stretching used in our experiments is non-injurious, as it engages subcutaneous tissue layers in shear within the animal’s usual range of motion. The U-shaped dose–response observed in our passive stretch model (30% stretch enhanced inhibition of tumor growth compared to both lower and higher amplitudes, Fig. 3) is consistent with a hormetic effect, which has been described in responses to a wide variety of stimuli [19, 27] This type of non-linear U-shaped dose–response also has been described in an animal exercise model in which moderate-intensity physical activity was more effective in inhibiting breast tumor growth, compared with both lower and higher intensity [15]. Notably, hormetic effects are typically in the 30–60% range, which is consistent with our results [7].
Accurate noninvasive measurement of tumor volume in vivo is important for monitoring the growth of tumors in longitudinal studies. In the orthotopic model, ultrasound was used to measure the volume of the tumor itself, whereas measurements with caliper included both the tumor and surrounding associated matrix. This finding is consistent with our ultrasound-measured volumes being approximately one-third of the caliper-measured volumes. Our finding that stretching reduces tumor volume via both methods by approximately the same amount (percentage) suggests that the effect of stretching not only reduces the volume of stroma around the tumor but also reduces the volume of cancer cells. Ultrasound measurements were particularly valuable in the transgenic model, where we were able to measure the effect of stretching at the very early stage of tumor formation (Fig. 2).

Conclusions
In summary, we have shown in several mouse mammary cancer models that moderate-amplitude, short-duration stretching reduces the local growth of tumors. These findings suggest that stretching could be beneficial in preventing local tumor growth, especially in patients, such as breast cancer survivors, who are at risk of recurrence. Our finding that stretching results in more benign tumor-associated collagen signatures in the MMTV-PYMT orthotopic model is especially important, as these signatures have been associated with improved prognosis in human breast cancer patients. Yoga is widely used and has been shown to be beneficial in breast cancer survivors, reducing pain and fatigue and increasing overall well-being [10, 11, 30]. To date, however, physical interventions such as yoga have not been tested for primary or secondary cancer prevention. Furthermore, the potential of mechanotherapies directed at the tumor stroma, in general, is understudied and mostly untapped [20]. The results of this study constitute an important step toward filling this significant knowledge gap.

Methods

Animals
The mouse strain FVB (Jackson Laboratory, Bar Harbor, ME) was used in all experiments. All animal procedures were performed under animal study protocols 22–1099 and were approved by the NIDCR Animal Care and Use Committee (ACUC). Mouse housing, care, and maintenance were provided by the NIDCR Veterinary Resources Core. Euthanasia was performed using CO2 inhalation followed by cervical dislocation. Anesthesia procedure was performed using 2–5% isoflurane.

Breeding and genotyping
Six-week-old male hemizygous MMTV-PYMT transgenic mice (strain #: 002374) (RRID:IMSR_JAX:002374) were purchased from the Jackson Laboratory and bred with 6-week-old FVB females (strain #: 001800) in house. A trio mating strategy (two females were housed with one male) was established to maximize the production of sufficient experimental animals. Mouse pups were weaned at 21 days of age and genotyped at 28 days of age through ear punching under isoflurane anesthesia.
DNA extraction and amplification were performed via an Extract-N-Amp™ Tissue PCR Kit (Sigma-Aldrich) to determine the genotype of the mice. The primer sequences (5′ → 3′) for the targeted gene and reference gene were as follows: Transgene (Forward: GGAAGCAAGTACTTCACAAGGG, Reverse: GGAAAGTCACTAGGAGCAGGG) and internal positive control (Forward: CAAATGTTGCTTGTCTGGTG, Reverse: GTCAGTCGAGTGCACAGTTT). Touch-down PCR was carried out in a Veriti™ 96-Well Fast Thermal Cycler (Thermo Fisher Scientific) with the following cycling profile: 3 min hold at 94 °C, 30 s at 94 °C for denaturing, 30 s at 65 °C for annealing, and 45 s at 72 °C for extension, followed by 9 cycles at decreasing annealing temperatures in decrements of 0.5 °C per cycle, then 30 s at 94 °C, 28 cycles of 30 s at 60 °C, 45 s at 72 °C, and a final extension at 72 °C for 10 min. Following amplification, the PCR products were loaded onto an E-Gel® 1.2% with SYBR Safe™ gel (Invitrogen) in the E-Gel™ Power Snap Electrophoresis System (Invitrogen) to identify the genotype of the samples.

Active Stretch
Before injection, the mice were randomized to the Stretch or No Stretch group for 10 min twice/day for three weeks. For the experiments to evaluate the collagen signature, the stretching procedure lasted only one week because the tumor size at this point was ideal for microscopic visualization, and the impact of the intervention was already foreseen. Animals were stretched as previously described [4] by lifting them at the base of the tail until they reached an ~ 45° angle to horizontal (Additional file 1A). With minimal habituation, all mice held this position comfortably for 10 min without struggling or vocalizing. Mice randomized to the No Stretch group were removed from their cages and placed on a table, but not restrained or lifted, for the same amount of time as those in the Stretch group.

Passive Stretch

Orthotopic model Passive Stretch procedure
Animals were anesthetized with 2% isoflurane, and the limbs were restrained using surgical tape while they were in the prone position. Stretching was achieved by elongating the forelimbs and hindlimbs longitudinally to a distance equivalent to 25% displacement (distance measured from toe to toe; Fig. 2A). No Stretch was achieved by restraining the limbs, but no extension was applied.

Transgenic model Passive Stretch procedure
Animals were anesthetized with 2% isoflurane, and their limbs were subsequently restrained in the supine position via surgical tape. Stretching was achieved by extending the forelimbs on the transverse l axes (distance measured from elbow to elbow, Fig. 4A), reaching ~ 60% differential displacement between the resting and stretch positions. No Stretch was performed by restraining the limbs, but no extension was applied.

Cell culture
Mouse Met-1 cells (Met-1(fvb2) (RRID:CVCL_U373), a highly metastatic tumor cell line derived from the original MMTV-PYMT transgenic animal [25], were kindly donated by Dr. Kent Hunter from the National Cancer Institute, Bethesda MD. Met-1 cell lines were cultured in DMEM supplemented with 10% FBS, 1% penicillin and streptomycin, and 1% L-glutamine (Gibco) and maintained at 37 °C with 5% CO2. The cell line was tested for Mycoplasma using the MycoAlert PLUS Mycoplasma Detection Kit (Lonza) upon receipt.

Cell line injection
Six-week-old FVB mice underwent bilateral tumor cell injection. Single tumor cell suspensions (Met-1 or GFP-Met-1) were prepared in a mixture containing 40% Geltrex (Thermo Fisher, SC) and 60% DMEM. A total of 0.020 × 106 tumor cells in 100 µl were injected subcutaneously, bilaterally on the back, at the third mammary fat pad.

Generation of stably expressing GFP-Met-1 cells
Following 1 passage of monolayer culture, 5 × 105 Met-1 cells were seeded in 24-well plates and allowed to attach overnight. The wells were then rinsed once with PBS, and lentiviral transduction medium was added as follows: Ham’s F-12 K (Kaighn’s) medium (Thermo Fisher SC) with 10% FBS, TransDux and MAX Enhancer (SBi, system bioscience) (1:10 ratio), per/400 ml of culture media to each well, and prepacked LL110VA-1 pLL-CMV-RFP-T2A-Puro (Lenti-Labeler™ virus) viral particles > 2 × 106 IFUs (SBi, system bioscience). After 72 h of incubation at 37 °C, the transduction media was removed, and the cells were harvested, washed by centrifugation, and resuspended in F-12 K media supplemented with 1–10 μg/mL puromycin. The transduced cells were expanded, harvested, and stored in cryovials with freezing medium (10% DMSO in FBS) in a liquid nitrogen cell mixture until further use.

Tumor volume measurement
Tumors were measured weekly using two approaches: ultrasound and caliper. For ultrasound measurement, we performed 3D ultrasound using a Visual Sonics (RRID:SCR_022152) ultrasound scanner (15 MHz) under isoflurane anesthesia. The transducer was stabilized with a clamp perpendicular to the skin of the back, and a cine-recording from the tumor area was recorded while the ultrasound probe was automatically swept across the skin at a constant rate. The tumor volume was measured as illustrated in Additional file 1B and Additional file 2, and Vevolab software was used for 3D reconstruction. For caliper measurements, individual tumor volume was calculated as π/6 × L × W2 [35]. The total tumor burden was calculated as the average tumor volume on the right and left sides.

Three-dimensional spheroid culture model
In-depth protocols for this approach have already been published [28]. Briefly, an ultra-low attachment V-bottom (or U-bottom) 96-well plate (Costar from Corning) was used to seed Met-1 cells at a density of 4000/100 µL cells per well. The plate was then centrifuged for 3 min at 250 × g at room temperature to allow the cells to form compact spheroids via cell adhesion. After centrifugation, a reduced growth factor basement membrane matrix (Geltrex, Thermo Fisher) was added at a final concentration of 5% (100 µL per well). The plate was placed in a tissue culture incubator for at least 48 h to promote the formation of a basement membrane around the spheroid. The spheroids were then rinsed three times in PBS (phosphate-buffered saline, Gibco) before being seeded in 2.35 mg/mL rat-tail collagen I gels. Stretch chambers (Strex SC-0040, San Diego California, USA) (RRID:SCR_025869) were used to polymerize the gels, and small rectangular pieces of sponges were attached to the walls to enhance gel adherence. After the gels were polymerized at 37 °C for 30 min, serum containing DMEM supplemented with 10% FBS was added. Preglycated collagen gel solutions were made by mixing Type I collagen (Corning, Product Number 354249) and D-()-ribose (Sigma-Aldrich, Product Number R9629) to a concentration of 250 mM prior to gelation [1] and incubating at 5 °C for 6 or 12 days prior to spheroid preparation. A Strex STB-100–04 manual static stretching system was used to stretch the gels to achieve 23% strain. The gels were stretched for ten minutes twice/day for 3 days.

Crown area and cell migration analysis
ImageJ software (RRID:SCR_003070), a publicly accessible Java-based program developed by the National Institutes of Health (NIH, Bethesda, MD), was used to analyze the crown area of spheroids, maximum distance traveled by migrating cells, and total number of single-cell migrations on day three of stretching. For the crown area analysis, the target image was opened in ImageJ and converted to an 8-bit format. An investigator who was subsequently blinded to the study conditions (Stretch vs. No Stretch) created a binary image by applying thresholding through the following: image/adjust/threshold. The low and high threshold values were adjusted to effectively isolate the desired features, a step that requires subjective judgment based on the image’s content. After suitable thresholds were determined, a binary image was obtained. To analyze the identified objects within this binary image, we used the analyze particle tool, which provided essential data for subsequent analysis and interpretation of the identified object.
The calculation of the maximum distance traveled by migratory cells involved measuring the distance from the edge of the spheroid to the farthest point reached by the migratory cells (the sixteen cells that were farthest from the edge of the spheroid were measured in all the spheroids). The total number of individual cells that migrated away from the spheroid was also manually counted.

Gene expression analysis
The mRNA level in the tumor tissue was assessed via quantitative reverse transcription polymerase chain reaction (qRT‒PCR). To isolate total RNA, 1 mL of TRIzol® Reagent (Invitrogen) and zirconium oxide beads (0.15 mm, Next, Advance, Inc., NY) were added to frozen samples in 2.0 mL screw cap microcentrifuge tubes. The samples were subsequently homogenized on a FastPrep®−24 homogenizer (MP Biochemicals) at a speed of 6.0 m/s for 80 s. The RNA in the tissue lysate was then separated from the DNA by chloroform, precipitated with 70% ethanol, and then purified with a PureLink™ RNA Mini Kit (Invitrogen) according to the manufacturer’s instructions. After the quantity and purity of the isolated RNA were determined via NanoDrop One (Thermo Fisher Scientific), the RNA samples were either stored at − 80 °C until use or immediately transcribed into cDNA via the SuperScript™ III First-Strand Synthesis System (Invitrogen) for further qRT‒PCR.
We selected 10 target genes related to the tumor extracellular matrix as well as three candidate reference genes and analyzed their expression by using commercially available Predesigned TaqMan Gene Expression Assays (Thermo Fisher Scientific) on Quant Studio™ 5 Real-Time PCR System (Thermo Fisher Scientific) (RRID:SCR_020240). Briefly, a total volume of 20 μL of reaction mixture consisting of 10 μL of TaqMan® Fast Advanced Master Mix (Thermo Fisher Scientific), 1 μL of a TaqMan® probe and primer set, and 2 μL of cDNA was loaded into each well of a 96-well plate. Three technical replicates were run for each biological replicate. Given that expression consistency may be impacted by different experimental conditions, the stability of three candidate reference genes, namely, GAPDH, Bm2, and 18 s, was evaluated using a web-based tool RefFinder (http://www.heartcure.com.au/reffinder/). GAPDH, which presented the greatest stability across the experimental treatments, was selected as the reference gene to normalize the expression of the target genes Col5a1, Col6a2, αSMA, TGFB1, Piezo1, Connexin 43, DDR1, DDR2, and Loxl3 in the subsequent 2−ΔΔCT analysis. The relative expression of a target gene is presented as a geometric mean with a 95% confidence interval (CI).

Confocal microscopy image acquisition and analysis
All imaging and image analyses were performed by an investigator blinded to the study conditions (Stretch vs. No Stretch). Tumors from 9- to 12-week-old mice (N = 16) were excised by carefully dissecting the subcutaneous tissue plane, allowing the tumor to remain attached to the skin (Additional file 1A). The entire excised tissue block was fixed in 4% paraformaldehyde overnight. The tissue was placed facing down and immersed in PBS on a glass bottom MatTek dish (MatTek Corporation). Tumors and the surrounding collagen matrix were imaged using a Nikon A1R HD MP system (RRID:SCR_020319). (Nikon Instruments, Melville, NY) together with a Coherent Chameleon Vision II two-photon laser (Coherent, Santa Clara, CA). Images were acquired using a 25 × (1.05 N.A.) silicone oil immersion objective. Simultaneous second harmonic generation (SHG) and EGFP imaging were achieved using 820 nm excitation with the emission wavelengths set to 400–450 and 500–550 nm, respectively. Since the tumor cells were GFP positive, the outer edges of the tumor were marked and identified via epifluorescence microscopy. Image tiling based on the tumor edges was then captured using a 1.5 × zoom in resonant mode (7 frames per second) at 1024 × 1024-pixel resolution. A motorized Z-piezo stage (Mad City Labs, Madison, WI) was used to rapidly capture Z-stacks every 1 micron over a Z-distance of 100 microns. NIS-Elements (Nikon) controlled all equipment, and the AI denoise algorithm was used to process images after acquisition. Data was then subjected to maximum intensity projection prior to further analysis. The SHG file was prepared in Imaris (RRID:SCR_007370) by finding the middle of the tumor. This was accomplished by observing which slices of the GFP/Met-1 cells were first observed and which slices of the GFP/Met-1 cells were last observed. The middle of these values is the slice that is determined to be the middle of the tumor. Once the image is acquired in Imaris, it is opened in ImageJ to subtract the background and accentuate the collagen fibers. The image is first processed through CT-FIRE, which traces individual fibers. The output of CT-FIRE is used as an input to CurveAlign, which analyzes fibers relative to the tumor boundary [24]. All collagen fibers with angles between x–y° were classified as TACS-2, and fibers with angles between a-b° were classified as TACS-3.

Spatial transcriptomics sample preparation
Twenty females FVB that were orthotopically injected with the Met-1 cell line were randomly assigned to No Stretch or Active Stretch, twice a day for 10 min for 2 weeks. At the end of the experiment, tumors were harvested, fixed in 4% PFA for 48 h, then paraffin embedded. A tissue microarray (TMA) paraffin block was prepared containing 1 mm sections from each of the tumors. Then, a 5-mm FFPE tissue section was obtained from the TMA block and applied onto the Xenium slide (10X Genomics, Pleasanton, CA) within the imaging area, following the manufacturer’s instructions. Briefly, this section was deparaffinized and permeabilized to make the mRNA accessible. The mRNAs were targeted by a panel of custom designed probes (n = 100). Probe hybridization was performed overnight at 50 °C with a probe concentration of 10 nM. After a stringency wash to remove un-hybridized probes, the probes were ligated at 37 °C for two hours, during which a rolling circle amplification (RCA) was performed. The circularized probes were then enzymatically amplified, producing multiple copies of the gene-specific barcode for each RNA binding event. After washing, background fluorescence was chemically quenched. The xenium slide was then placed into an imaging cassette for loading onto the xenium analyzer instrument.

Gene panel design
The Xenium in Situ technology (10X Genomics, Pleasanton, CA) employs targeted panels to detect gene expression; in this experiment we used a custom designed panel containing 100 genes, which were chosen based on previous studies using single cell transcriptomics on the MMTV PYMT model to study tumor microenvironment cell composition and distribution [3, 17, 17, 18], including the use of conserved markers to identify cell populations such as epithelial cells (luminal/basal), macrophages, T cells, NK cells, endothelial cells, and fibroblasts [18].

Spatial transcriptomics analysis
Xenium data was analyzed with the Seurat R package. Selected tissue microarray cores were chosen for downstream analysis using the Crop() function. Group metadata was added to cell-based coordinate locations, and cells featuring less than 100 Xenium counts were removed from subsequent analysis. The total number of cells remaining was 19,9087. Expression levels were normalized using the SCTransform() function, and PCA, clustering, and UMAP embeddings were computed using the first 30 dimensions and default parameters. Optimal clustering resolution was determined by the Clustree R package.
To identify cell populations, we performed clustering and cell type annotation based on canonical markers and then downsampled each condition to have an equal number of cells (66,013 cells per group). We computed the total number of cells per cell type and per tissue Sect. (10 Stretch, 10 No Stretch) and analyzed this table by the scCODA method, which aims to identify changes in compositional parameters as previously described [6].

Ultrasound cine-recording of tissue displacement during passive stretching
In order to visualize the mechanical impact of transverse stretching on the tissue layers surrounding subcutaneously implanted mammary tumors, we performed an ultrasound cine-recording under anesthesia while passively stretching the tissues, with the axis of stretching parallel to the skin. For the recording, stretching was cyclical, rather than static, to better demonstrate the tissue displacement visually.

Statistical analyses
Mixed model repeated measures analyses were used to compare stretch and no stretch groups on tumor volume, as measured by the caliper and ultrasound methods previously described. Group (Stretch vs No Stretch) and time (weeks) were fixed factors in the model. In experiments in which the animals were randomized to stretch and not stretch within a batch of animals, the batch was included as a random effect in the model, which represented replicates of the experiment. Tumor volume was log transformed prior to analysis to satisfy homogeneity of variance assumptions. Means presented for tumor volume are geometric means derived from the log transformed analysis. Simple effects (i.e., time-specific group comparisons) were performed using partial F tests. Pearson’s r was used to quantify the linear association between caliper and ultrasound tumor volume measurements. Tumor-associated collagen signatures (TACS) were compared between stretch and No Stretch groups in a subset of animals (N = 8) to investigate the impact of stretching on the extracellular matrix. The mean percentage of TACS 2 collagen fibers oriented parallel (< 15-degree angle) and TACS 3 collagen fibers aligned perpendicular (> 75-degree angle) to the boundary of the tumors, and their ratios were compared between the stretch and no stretch groups using t tests. Ratios were log transformed prior to analysis, and the mean ratios presented represent geometric means derived from the log transformed analysis. Mixed models were also used to compare stretch and no stretch conditions on outcomes from the in vitro model, which included gel as an additional random factor. For these analyses, the gel represented the experimental unit, which was randomized, and the spheroids were subsamples nested within the gel. Statistical analyses were performed using SAS Statistical Software V 9.4 (SAS Institute, Cary, NC) (RRID:SCR_008567). Statistical significance was set at p < 0.05.

Supplementary Information