Rapid cancer cell perineural invasion utilizes amoeboid migration

Significance

Cancer cells invade and travel along nerves in a process termed perineural invasion, which is a sign of aggressive disease. To better understand this process, a mouse model was used to select cancer cells that travel rapidly along nerves. These cancer cells exhibited an amoeboid mode of migration, which is marked by a round, blebbed, fast moving, pliable cell phenotype. This migration mode can be modulated pharmacologically. This work shows that cancer cells can modulate their migration mode to enable rapid nerve invasion, and that this mechanism may be targeted therapeutically.

Abstract

The invasion of nerves by cancer cells, or perineural invasion (PNI), is potentiated by the nerve microenvironment and is associated with adverse clinical outcomes. However, the cancer cell characteristics that enable PNI are poorly defined. Here, we generated cell lines enriched for a rapid neuroinvasive phenotype by serially passaging pancreatic cancer cells in a murine sciatic nerve model of PNI. Cancer cells isolated from the leading edge of nerve invasion showed a progressively increasing nerve invasion velocity with higher passage number. Transcriptome analysis revealed an upregulation of proteins involving the plasma membrane, cell leading edge, and cell movement in the leading neuroinvasive cells. Leading cells progressively became round and blebbed, lost focal adhesions and filipodia, and transitioned from a mesenchymal to amoeboid phenotype. Leading cells acquired an increased ability to migrate through microchannel constrictions and associated more with dorsal root ganglia than nonleading cells. ROCK inhibition reverted leading cells from an amoeboid to mesenchymal phenotype, reduced migration through microchannel constrictions, reduced neurite association, and reduced PNI in a murine sciatic nerve model. Cancer cells with rapid PNI exhibit an amoeboid phenotype, highlighting the plasticity of cancer migration mode in enabling rapid nerve invasion.

Perineural invasion (PNI) is characterized by the histologic identification of cancer cells invading along or around nerves and is an adverse prognostic factor for multiple cancer types that is associated with pain, numbness, and nerve paralysis (1–4). Cancer invasion along nerves allows for insidious and extensive spread along neural pathways and represents a mechanism of cancer spread that is distinct from local invasion, nodal metastases, or distant metastases. Tumors arising from highly innervated organs are more often associated with PNI, which occurs commonly in cancers of the pancreas, head and neck mucosa, prostate, and gastrointestinal tract, among others. The histologic identification of PNI after surgical resection is a clinically ominous finding that is associated with an elevated risk of cancer recurrence and decreased patient survival rates (2–5). Pancreatic cancer, in which PNI is highly prevalent, is a highly deadly disease with a 5-y overall survival rate of just 11% (6).

PNI is driven by reciprocal interactions between cancer cells and the nerve microenvironment (2–5, 7, 8). Glial-derived neurotrophic factor (GDNF) is released by Schwann cells in response to neve injury and binds to RET and GFRα receptors on cancer cells to induce directional cell migration toward the nerve and subsequent cancer invasion through Cdc42 signaling (9–11). In addition, Schwann cells become activated, as they do following nerve injury, enabling them to disperse and recruit cancer cells to nerves through direct cell contact (12) and also create pathways to promote cancer invasion (13). Schwann cells also release CCL2 to recruit inflammatory monocytes to sites of PNI, which then differentiate to macrophages and enhance cancer invasion through the release of cathepsin B (14). The nerve microenvironment is therefore a very active participant in enabling PNI.

Cancer cell factors are also necessary for establishing a neuroinvasive phenotype, as some cancer types clearly exhibit a higher predisposition for PNI than others (2, 8). However, the cancer cell determinants that facilitate rapid PNI are poorly understood. Prior studies have explored markers of PNI by comparing different pancreatic cancer cell lines with varying predilections for PNI and identified two candidate markers: CD74 and synuclein-gamma (15, 16). However, these studies cannot account for the innumerable innate differences in the many cell lines. Another study used an ex vivo nerve invasion assay to select pancreatic cancer cells for cDNA microarray, identifying KIF14 and ARHGDI-beta as candidate markers (17).

Here, we sought to define the intrinsic cancer cell properties that enable certain cancer cells to migrate more rapidly along nerves in vivo. To study this phenotype, we used a murine model of PNI to select for a series of increasingly neuroinvasive cancer cells over multiple serial passages. After validating enriched neuroinvasive speed in the selected leading cells, we examined the transcriptomic signatures and cellular characteristics of these cancer cells, which had gained a predilection toward rapid nerve invasion. Our findings highlight the importance of alterations in cell migration gene expression and of mesenchymal to amoeboid transition in accelerating PNI. This study defines cancer migration mode as a determinant of neuroinvasiveness and offer insights into therapeutic strategies for this aggressive type of cancer invasion.

Results

Nerve Invasion Velocity Increases with Serial Cancer Cell Passages in a Murine Sciatic Nerve Model.

We generated a cancer cell phenotype enriched in its ability to invade nerves by performing serial injections of a red fluorescent protein (RFP)-labeled human pancreatic cancer cell line (MiaPaCa-2) into the distal sciatic nerves of Nu/J mice (Fig. 1A) (18). Mice were killed after the majority had developed impaired sciatic nerve function, as measured by a sciatic neurological score of 2 or less. This scoring system assesses hind limb motor response to limb extension (4 = normal and 1 = full paralysis, Fig. 1B) (9, 12, 19). The injected sciatic nerves were harvested and proximal nerve invasion toward the spinal cord was assessed by measuring the length of the thickened portion of sciatic nerve (Fig. 1C). Nerve invasion was not identified in the distal direction. Histologic examination confirmed cancer infiltration of the thickened sciatic nerves (Fig. 1D). The most proximal and the most distal (primary tumor) portions of the invaded nerve were surgically isolated, digested, and fluorescent cancer cells were isolated by fluorescence-associated cell sorting (FACS) (Fig. 1E). RFP-labeled cancer cells collected from the proximal region of the invaded nerve near the spinal cord were termed leaders (Lead-1), while those from the distal region of nerve at the injection site were termed laggers (Lag-1). Isolated cells were collected for RNA extraction, cryopreservation, and expansion in cell culture.

Fig. 1.

Nerve invasion velocity increases with serial cancer cell passages in a murine sciatic nerve model. (A) Schematic diagram showing serial injections of RFP-labeled MiaPaCa-2 cancer cells into distal sciatic nerves of Nu/J mice to generate a neuroinvasive cancer cell phenotype. After injection, RFP-labeled cancer cells at the proximal and distal portions of the sciatic nerve were sorted, expanded in culture, then reinjected into the sciatic nerves of new mice. This process was repeated 4 times, yielding a series of new cells labeled by passage number through the sciatic nerve. (B) Representative images of murine hind limbs after PBS or cancer injection into the right sciatic nerve. Mice were killed after the majority had developed impaired sciatic nerve function as measured by a sciatic neurological score. (C) Surgical isolation of injected and noninjected sciatic nerve. Cancer cells at the proximal and distal ends of the nerve were termed Lead and Lag cells, respectively. (D) H&E section of sciatic nerves injected and noninjected with cancer cells. (Scale bar, 1,000 µm.) (E). FACS isolation of RFP-labeled cancer cells from proximal and distal regions of the invaded nerve after harvest and digestion. (F) Velocity of sciatic nerve invasion per week by Lead cells. (G) Length of sciatic nerve invasion 4 wk after injection of Lead-4 and Lag-1 cells. (H) Functional assessment of hind limb function after injection of Lead-4 and Lag-1 cells, using a sciatic neurological score that assesses hind limb motor response to limb extension. Comparisons performed with two-tailed, Student’s t test.

Lead-1 cells grown in culture were subsequently reinjected into the sciatic nerves of new mice. This process was serially repeated four times, yielding a series of new cells labeled by passage number through the sciatic nerve (Lag-1, Lead-1, Lead-2, Lead-3, Lead-4) (Fig. 1A). The velocity of sciatic nerve invasion per week by the Lead cells progressively increased over each of the three serial passages before leveling off with the fourth passage (Fig. 1F). The sciatic nerves of Nu/J mice were injected with Lead-4 or Lag-1 cells for validation. After 4 wk, Lead-4 cells had invaded a significantly longer distance along the sciatic nerve as compared to Lag-1 cells (mean 1.95 cm vs. 1.5 cm, P = 0.03) (Fig. 1G). Functional assessment demonstrated that these mice injected with Lead-4 cells experienced more limb paralysis as compared to mice injected with Lag-1 cells at 4 wk (Fig. 1H). Parental MiaPaCa-2, Lag-1, and Lead1–4 cell proliferation in vitro was assessed using MTT assay. There were no significant differences noted over 4 d, with Lead-4 showing a slightly lower proliferation on day 5 as compared to the other cell lines (SI Appendix, Fig. S1A). These results control for the possibility of altered Lead-4 proliferation as a confounding factor that could increase PNI length. Lead-4 will be referred to as “Lead” cells, while Lag-1 will be referred to as “Lag” cells.

Lead Cell Transcriptome Reveals an Upregulation of Cell Migration Genes.

To determine whether Lead and Lag have changes in gene expression associated with an increasing neuroinvasive phenotype, we performed RNA sequencing on the initial lagging RFP-MiaPaCa-2 cells (Lag-1) and the leading cells isolated after each passage in the sciatic nerve model. Genes with progressively increasing expression with each passage through the nerve were identified (510 in total), and gene ontology (GO) pathway analysis was performed (Fig. 2A). The top 10 up-regulated GO pathways were identified, with three involving the plasma membrane, three involving the cell leading edge, and one involving cellular component movement (Fig. 2B). A heat map of gene expression changes and genes overlapping multiple categories are reported (Fig. 2C). Importantly, we identified a progressive upregulation in genes that regulate the actomyosin contractility, which drive an amoeboid type of cellular migration, including rho associated protein kinase 2 (ROCK2), paxillin (PXN), vinculin (VCL), podocalyxin-like protein 1 (PODXL), gelsolin (GSN), and myosin X (MYO10) (Fig. 2C, asterisks).

Fig. 2.

Lead cell transcriptome reveals an upregulation of cell migration genes. (A) Schematic of gene expression analysis of Lead and Lag cells. RNA sequencing of 21,416 genes identifies 510 progressively up-regulated genes with each passage, which are then examined with GO pathway analysis. (B) The top 10 biological process GO terms that are progressively up-regulated with each neural passage. (C) Heat map of selected genes that are progressively up-regulated with each passage. Venn diagram indicates overlap of genes involved in multiple pathways. Asterisk represents genes that regulate the actomyosin contractility, which drive an amoeboid type of cellular migration.

We also identified the 90 most down-regulated genes with increasing passage. These were assessed using Enrichr for KEGG and GO pathway analyses (SI Appendix, Fig. S1B). KEGG pathways involving purine metabolism, pyrmidine metabolism, mTOR signaling, and taurine metabolism were identified. There were no pathways involving cell migration identified.

Increasingly Neuroinvasive Cells Undergo Progressive Mesenchymal to Amoeboid Transition.

Leading cells harvested after each passage in the murine PNI model were cultured and compared morphologically in vitro with Lag-1 cells. Lag cells were heterogeneous and tended to have an elongated, spindle appearance. In contrast, Lead-1-4 cells exhibited progressively increasing roundness with higher neural passages (roundness index 0.67 vs 0.42; P < 0.0001) (Fig. 3A). Immunofluorescence staining confirmed the increased expression of the actomyosin contractility proteins GSN, PODXL, EZR, PXN, and ROCK1 in Lead as compared to Lag cells that was consistent with the RNA sequencing results (P < 0.0001) (Fig. 3 B and C).

Fig. 3.

Increasingly neuroinvasive cells undergo progressive mesenchymal to amoeboid transition. (A) Morphological comparison of MiaPaCa-2, Lag and Lead cells expressing RFP. Roundness index of Lead and Lag cells shows increasing roundness with higher passages, suggesting a mesenchymal-to-amoeboid transition. Experiment performed twice. (B) Immunofluorescence staining and mean fluorescence density of the actomyosin contractility proteins GSN, PODXL, EZR, PXN, and ROCK1 in Lead and Lag cells. (C) Heat map of actomyosin contractility migration and amoeboid migration gene expression of cells through increasing passage number. Comparisons performed with two-tailed, Student’s t test. Experiment performed twice. (D) Plot of HIC-5 to PXN ratio of Lead and Lag cells based on RNA-seq data. A high HIC-5 to PXN ratio is associated with a mesenchymal phenotype, while a low ratio is associated with an amoeboid phenotype. Comparisons performed with two-tailed, Student’s t test.

An upregulation in the Rho/ROCK signaling pathway, increased actomyosin protein expression, and a shift toward a rounder cell morphology with higher neural passages suggests an alteration in the migratory phenotype of the cells from a mesenchymal mode in the Lag cells, to an amoeboid mode in the Lead cells. This change suggests that these cell lines have progressively undergone mesenchymal-to-amoeboid transition (MAT) with increasing passage. The genomic expression profile of Lead and Lag cells shows an upregulation of ROCK1 and ROCK2 (P < 0.05 for both), drivers of amoeboid cellular migration (Fig. 3C). There was also a trend toward an upregulation of other genes associated with amoeboid migration (EPHA2, MYH9, STMN1, LIMK1). There was no significant change in RhoA, RhoC, or Rac1 expression, drivers of mesenchymal migration.

Increased paxillin and decreased HIC-5 leads to an amoeboid state, while decreased paxillin and increased HIC-5 leads to a mesenchymal state (20). The ratio of HIC-5 to paxillin predicts cancer cell morphology and phenotypic plasticity. Cells that remain stable in an amoeboid state exhibit a low HIC-5 to paxillin expression ratio, while cells that exhibit high plasticity and shift spontaneously to a mesenchymal morphology have high HIC-5 to paxillin ratios (21). RNA-Seq data revealed a progressive increase in PXN with higher neural passages, with lesser alterations in HIC-5. The HIC-5/PXN ratio was >0.5 in Lag-1 cells but dropped to <0.5 in Lead 1–4 cells (Fig. 3D). These results suggest a progressively increasing mesenchymal to amoeboid transition from Lag-1 to Lead-4.

We also compared the cellular protein distribution of Lead and Lag cells with immunofluorescence microscopy for actin, ezrin, and paxillin. In Lag cells, actin (phalloidin staining) was organized mostly as stress fibers. In contrast, Lead cells were mostly characterized by cortical actin, with high expression at the cell membrane and along bleb-like structures (Fig. 4A). In Lag cells, ezrin (EZR) was expressed in cellular protrusions, filopodia, and at cell–cell adhesions. In Lead cells, EZR was observed at cell membrane blebs and showed greater diffuse expression overall, consistent with the RNA sequencing results. Using EZR staining as a cell membrane marker of filipodia, we counted a significantly a higher percentage of Lag cells with filopodia as compared to Lead cells (93% vs. 36%; P < 0.0001). In addition, there was a significantly higher percentage of Lead cells forming cell membrane blebs as compared to Lag cells (76% vs. 26%; P < 0.0001) (Fig. 4A).

Fig. 4.

Lead and Lag cell phenotype. (A) Immunofluorescence microscopy and confocal microscopy staining of phalloidin (actin), ezrin, paxillin, and vinculin in Lead and Lag cells. Right graphs show percent cells with stress fibers, filopodia, blebs, focal adhesions and the length of focal adhesions based on phalloidin, ezrin, and paxillin staining, respectively. Comparisons performed with Fischer’s exact test (stress fibers, filipodia, blebs) or two-tailed, Student’s t test (focal adhesions). Experiment performed twice. (B) A RhoA fluorescence resonance energy transfer (FRET) biosensor demonstrates increased RhoA activity in Lead as compared to Lag cells. Comparisons performed with two-tailed, Student’s t test. Experiment performed twice. (C) AFM was used to assess the elasticity of the Lead and Lag cells by measuring the response to a force applied on cells by a cantilever. (D) Stiffness map of Lead and Lag cells measured by AFM. (E) Lead cells were significantly less stiff and more deformable as compared to the Lag cells. Comparisons performed with two-tailed, Student’s t test. Experiment performed three times.

In Lag cells, PXN staining showed numerous focal adhesions, while in Lead cells there were fewer focal adhesions (14.0 vs. 1.4 focal adhesions per cell; P < 0.0001). These findings were validated on confocal microscopy, where Lag cells also demonstrated longer focal adhesions as compared with Lead cells. In Lead cells, PXN was expressed diffusely throughout the cell’s cytoplasm with greater expression in Lead than Lag cells but clearly showed greater expression in focal adhesions in the Lag cells. VCL was also expressed more in the cytoplasm of Lead cells as compared to Lag cells on confocal microscopy, but not within focal adhesions (Fig. 4A). Confocal microscopy also validated a dramatically higher actin stress fiber content in Lag cells as compared with Lead cells on phalloidin staining. Collectively, these alterations in actin organization, cell–cell adhesion, filopodia, focal adhesions, and membrane blebbing suggest that a mesenchymal–amoeboid transition has occurred from Lag to Lead cells.

A RhoA fluorescence resonance energy transfer (FRET) biosensor was used to demonstrate increased RhoA activity in Lead as compared to Lag cells (Fig. 4B). The ratio of phospho-cofilin to total cofilin, a downstream marker of RhoA activity, was also found to be increased by 2.3 fold in Lead as compared to Lag cells on immunoblot (SI Appendix, Fig. S1C). Amoeboid cells are more deformable and pliable than mesenchymal cells (22). Atomic force microscopy (AFM) was used to measure the elasticity of the Lead cells compared to the Lag cells (Fig. 4C). In response to a force applied by a cantilever on cells plated on glass dishes, Lead cells were significantly less stiff (0.40 kPa; SD = 0.19) and more deformable as compared to the Lag cells (0.95 kPa; SD = 0.35, P < 0.0001) (Fig. 4 D and E). This finding also supports an amoeboid phenotype for Lead cells as compared to a mesenchymal phenotype for Lag cells.

Lead vs. Lag Cells Exhibit Altered Migration Dynamics in 2D vs. 3D.

Efficient cellular migration on a two dimensional (2D) surface requires the expression of adhesion proteins to anchor the cells to the surface and propel them forward (22). Cells that express few adhesion proteins, such as amoeboid cells, migrate less effectively in 2D as compared with cells expressing more adhesive proteins, such as mesenchymal cells. We tracked the 2D migration patterns of the Lag and Lead cells seeded on a glass plate over 24 h (Fig. 5A). Time-lapse cell tracking over 8 h shows that Lag cells migrated further distances than Lead cells (189.1 ± 115.7 vs. 111.0 ± 70.9 µm, mean ± SD, P < 0.0001), at higher mean velocity (0.52 ± 0.24 vs. 0.17 ± 0.16 µm/min; P < 0.0001) and had higher directional persistence as measured by the straightness index of migration paths (0.60 ± 0.25 vs. 0.27 ± 0.23; P < 0.0001) (Fig. 5 B and C).

Fig. 5.

Lead and Lag cells exhibit altered migration dynamics in 2D vs. 3D. (A) 2D migration patterns of the Lag and Lead cells seeded on a glass plate over 24 h. (B) Time-lapse cell tracking of Lead and Lag cells migrating in 2D over 8 h. (C) Comparison of distance migrated, velocity, and directionality of Lead and Lag cell migration in 2D. Comparisons performed with two-tailed, Student’s t test. Experiment performed three times. (D) Representative time-lapse images and velocity of Lead and Lag cells seeded in 12-µm-wide, 10-µm-high, 3D microchannels. Comparisons performed with two-tailed, Student’s t test. (E, Left) Representative time-lapse images of Lead and Lag cells seeded in 12-µm-wide, 10-µm-high, 3D microchannels which contain 8-µm-wide, 5-µm-long, constrictions (dT = hours:minutes). (Middle) Schematic diagram of permeation, the complete passing of cells through microchannel constrictions. (Right) Percent Lead and Lag cells with constriction permeation. Comparisons performed with Fisher’s exact test. (F, Left) Schematic diagram demonstrating Lead/Lag cells seeded in a 3D gel matrix in proximity to a murine dorsal root ganglion (DRG) explant. This assay assesses the ability of cancer cells to migrate through gel matrix to associate with nerves. (Middle) Representative fluorescence microscopy image of cancer cells associating with DRG. (Right) Percentage of Lead and Lag cells with neurite association. Comparisons performed with Fisher’s exact test. Experiment performed twice.

When cells are in a three-dimensional (3D) space, cell surface anchoring by adhesive proteins is less critical for migration. Amoeboid migration is enhanced in confined 3D spaces and less dependent on surface anchoring (23–25). Lead and Lag cells seeded in 12-µm-wide, 3D microchannels show a velocity advantage to the Lead as compared to Lag cells (0.35 µm/min vs. 0.12 µm/min, P = 0.02) (Fig. 5D and Movies S1 and S2). The fastest Lead cells showed a round, amoeboid morphology during microchannel migration, while Lag cells were generally mesenchymal.

Lead or Lag cells were next seeded in 12-µm-wide, 3D microchannels which contain 8-µm-wide, 5-µm-long, constrictions located every 150 µm (Fig. 5E and Movies S3 and S4). Migration of cells through the constrictions were quantified. Lead cells were more likely to pass completely through the constriction (permeation) as compared to Lag cells (55.7% vs. 40.2%; P = 0.0024). To assess the ability of these cancer cells to migrate over and through gel matrix and associate with nerves, Lead or Lag cells were seeded in a 3D gel matrix in proximity to a murine dorsal root ganglion (DRG) explant from a C57BL/6 mouse. Lead cells were more likely to invade through gel and associate with DRG neurites as compared to Lag cells (83.3% vs. 57%, P < 0.0001) (Fig. 5F).

These data indicate an enhanced 3D migration capability of Lead cells as compared to Lag cells. Lead cells are generally more round, deformable, and move rapidly in microchannels. Lead cells navigate more easily through microchannel constrictions and more readily invade through 3D gel matrix to associate with DRG neurites. Lag cells exhibited increased migration on 2D with stronger adhesion characteristics and a mesenchymal phenotype. These findings are consistent with a preferred amoeboid mode of migration by Lead cells, contrasting with a shift to a mesenchymal mode of migration by Lag cells, reflecting plasticity between these migration patterns.

ROCK Inhibitors Induce Amoeboid to Mesenchymal Transition in Lead Cells.

Rho-kinase (ROCK) is a driver of the amoeboid phenotype and amoeboid-type migration of cells (26). Both ROCK 1 and 2 were up-regulated in Lead cells as compared with Lag cells. Inhibition of ROCK has been previously shown to reverse the amoeboid cellular phenotype, causing an amoeboid to mesenchymal transition (AMT) (26). After treating Lead cells with the ROCK inhibitor Y27632 (25 µM), cells became elongated and spindle-shaped, with a decrease in roundness index from 0.73 to 0.54 (P < 0.0001) (Fig. 6A). There was no significant change in the morphology of Lag cells, which retained their spindle shape after treatment with Y27632.

Fig. 6.

ROCK inhibitors induce AMT in Lead cells. (A) Representative fluorescence microscopy images (Left) of Lead and Lag cells with and without exposure to a ROCK inhibitor. After exposure, round Lead cells become elongated and spindle-shaped, with a decrease in roundness index (Right). Comparisons performed with two-tailed, Student’s t test. Experiment performed twice. (B) Immunofluorescence staining of paxillin and ezrin in Lead and Lag cells with and without ROCK inhibitor treatment. Right graphs show focal adhesions per cell, and percent of cells with filopodia and blebs based on paxillin and ezrin staining, respectively. Comparisons performed with two-tailed, Student’s t test (focal adhesions) or Fisher’s exact test (filipodia, blebs). Experiment performed twice. Experiment performed twice. (C) Percent Lead cells that permeate through microchannel constrictions with and without ROCK inhibitor treatment. Comparisons performed with Fisher’s exact test. Experiment performed twice. (D) Percent Lead cells that associate with DRG neurites with and without ROCK inhibitor treatment. Comparisons performed with Fisher’s exact test. Experiment performed twice. (E) Length of murine sciatic nerve invasion after inoculation distally with MiaPaCa cancer cells then treated with ROCK inhibitor or control three times a week for 4 wk. Comparisons performed with two-tailed, Student’s t test. Experiment performed three times. (F, Top) Representative gross image of isolated murine sciatic nerves with and without ROCK inhibitor treatment. (Middle) Corresponding H&E sections. (Scale bar, 2,000 µm.) (Bottom) Enlarged inset showing cancer cells injected distally and invading proximally in the control but not the inhibitor.

Lead cells treated with or without Y27632 were examined by immunofluorescence microscopy for PXN and EZR expression. Y27632 increased the number of PXN-rich focal adhesions (8.0 vs. 2.4 focal adhesions per cell; P < 0.0001), increased the percentage of cells containing filopodia (54% vs. 33%; P = 0.007), and decreased the percentage of cells with membrane blebs (18% vs. 74%; P < 0.0001) (Fig. 6B). Collectively, these observations with ROCK inhibition are consistent with Lead cells undergoing an amoeboid to mesenchymal transition and resembling Lag cells.

Lead cells treated with or without Y27632 were seeded in 12-µm-wide microchannels with constrictions. Untreated Lead cells were more likely to pass through the constrictions as compared to Lead cells treated with Y27632 (52% vs. 39%; P = 0.0498) (Fig. 6C). Lead cells treated with Y27632 permeated the constrictions at approximately the same rate as Lag cells (Figs. 5D and 6C). When placed in a dish with a DRG in Matrigel, untreated Lead were significantly more likely to associate with DRG neurites than Lead cells exposed to Y27632 (83.3% vs. 52.3%; P < 0.0001) (Fig. 6D). These studies demonstrate that the phenotypic differences that Lead cells acquired with enhanced migration through 3D constrictions and association with nerves in a 3D matrix are lost when they are pharmacologically transitioned back to a mesenchymal state.

ROCK Inhibition Reduces PNI in a Murine Sciatic Nerve Model.

To assess whether ROCK inhibition could reduce the length of PNI in vivo, we treated a group of mice with sciatic nerve MiaPaCa-2 tumors with either a ROCK inhibitor or with control. After inoculation of the sciatic nerve with MiaPaCa-2 cells, mice were treated with 1 mg (female) to 1.25 mg (male) ROCK inhibitor Y27632 (5 mg/mL) via intraperitoneal injection three times a week for 4 wk. We noted that the mean length of sciatic nerve invasion proximal to the tumor mass was significantly reduced with ROCK inhibition as compared with controls (10.3 mm vs. 4.2 mm, P = 0.0263) (Fig. 6 E and F). We did not identify any significant differences in hindlimb function in these studies.

The direct effects of Y27632 were examined on the proliferation of Lead and Lag cells in vitro by MTT assay. There were no significant differences in cell proliferation between control untreated Lead or Lag cells as compared to those treated with 5, 10, or 20 µM of Y27632 (SI Appendix, Fig. S1D). These results suggest that ROCK inhibition does not have a direct impact on Lead or Lag cell proliferation, highlighting the impact of ROCK inhibition on altering migration mode as its predominant influence on Lead cells.

Human Cancer Cell Line Expression of a 6 Gene Signature.

We assessed RNA sequencing data from a panel of over 1,400 cell lines using the Cancer Cell Line Encyclopedia (CCLE) from the Broad Institute. The expression was assessed of the 6 genes up-regulated in Lead cells (Fig. 2C; ROCK2, PXN, GSN, PODXL, VCL, and MYO10). The expression level of each gene was scaled considering 100% equal to the maximum expression level of any cell line in the CCLE database. A composite score of all 6 genes was then summed together to create an overall index. The cell lines with the 24 highest and lowest sum scores were assessed (Fig. 7A).

Fig. 7.

Expression of Lead up-regulated proteins in human cancers. (A) RNA sequencing data from the Cancer Cell Line Encyclopedia (CCLE). The expression of 6 genes up-regulated in Lead cells was assessed and scaled considering 100% equal to the maximum expression level of any cell line in the database. Cell lines with the 24 highest and lowest composite sum scores are depicted. Yellow highlighted cell lines are those from cancers with a high propensity for neural invasion. (B) Immunofluorescence of Lead up-regulated proteins in human pancreatic adenocarcinoma specimens.

For the cell lines with the highest sum score of the 6 gene signature, half (12/24) of the cell lines (yellow highlight) originate from cancer types that are known to exhibit highly neurotrophic behavior. These include pancreatic ductal adenocarcinoma (PDAC, n = 4), cholangiocarcinoma (n = 5), esophageal adenocarcinoma (n = 1), and glioma (n = 2). For the cell lines with the lowest sum score of the 6 gene signature, none of the cell lines originate from malignancy types that exhibit neurotrophic behavior. Most of these cell lines are from hematogenous malignancies such as lymphoma, leukemia, and multiple myeloma.

We performed immunofluorescence microscopy of PXN, ROCK2, VCL, GSN, and PODXL on human pancreatic ductal adenocarcinoma surgical specimens (PDAC). As controls, nonneoplastic pancreatic tissue was used from pancreatic surgical specimens of patients with either benign serous cystadenoma (for PXN, ROCK2, VCL, and GSN) or ampullary tubulovillous adenoma (for PODXL). Costaining with S100 to identify nerves was performed. PDAC-specific staining of all these proteins was increased as compared with nonneoplastic pancreatic tissue, with a particularly marked increased staining of VCL and GSN in PDAC cells as compared with controls (Fig. 7B).

Discussion

PNI is a poorly understood mode of cancer spread in which cancer cells invade within and along nerves, often far beyond the appreciable primary tumor mass. There is a predilection for PNI by a variety of cancer types, including cancers of the pancreas, prostate, colon, head and neck, and skin among others. PNI causes pain, paresthesia, and paralysis, and is a poor prognostic factor that is associated with increased rates of cancer recurrence and diminished patient survival rates (2–4, 27–33). Recent studies have revealed the important role that the nerve microenvironment plays in enabling this adverse process (5, 7–10). However, the cancer-specific determinants which confer an ability of cancer cells to rapidly invade along nerves remain poorly defined. Here, our approach used an in vivo murine sciatic nerve model to generate an accelerated neuroinvasive pancreatic cancer phenotype for transcriptomic analysis.

Prior studies have established a key role for cancer cell migration in the mechanisms that underlie PNI. Nerves and Schwann cells release paracrine factors that function as chemoattractants to drive cancer cell migration (9, 10). In addition, the Rho GTPase Cdc42 steers cancer cells during chemotaxis and enables cancer cells to navigate successfully to nerves, mediated through nerve-released GDNF and Rearranged during Transfection (RET) receptor signaling (11). Directional cancer cell migration is a critical requirement of PNI. Here, we examined the genomic expression profiles of increasingly neuroinvasive pancreatic cancer cells to gain global insights into the cancer dependent factors that enable efficient and rapid PNI.

To select highly neuroinvasive cancer cells, we utilized a murine sciatic PNI model that has been optimized (9–13, 18, 19). Fluorescent labeling of MiaPaCa-2 cells allowed for serial injections and flow cytometry isolation of the furthest invading cells. MiaPaCa2 cells exhibit typical features of mesenchymal cells including a lack of E-cadherin and N-cadherin, expression of Snail and Vimentin, and a spindle cell morphology (34–36). With each passage, hind limb paralysis occurred faster, indicating an enrichment of traits adapted for rapid nerve invasion. A comparison of gene expression profiles revealed an upregulation of proteins involved in migration by leading cells, including proteins involved in the cell membrane, cell leading edge formation and cellular movement. With higher passages the neuroinvasive cells shifted from a mesenchymal to an amoeboid gene expression profile. ROCK1 and ROCK2, promotors of amoeboid migration (37, 38), were both up-regulated. This finding is consistent with a prior study demonstrating that RhoA controls bleb-based migration by confined cells with dorsoventral polarity (39). In addition, the upregulation of PODXL by Lead cells is also consistent with a prior study showing that PODXL modulates pancreatic cancer cell migration and metastasis through microtubule and Src-dependent pathways (40).

Lead cells became more amoeboid in their morphology with a progressively rounder shape and increased membrane blebbing. These cells lost mesenchymal characteristics of actin stress fibers, filipodia, and focal adhesions. Lead cell deformability increased as measured by AFM. Lead cells exhibited more rapid migration through 3D microchannels with constrictions, while lagging cells exhibited more rapid migration in 2D. These characteristics are all consistent with leading cell transition to an amoeboid mode of migration. Cancer cell invasion and metastasis are dependent on the ability of cancer cells to adapt to changing 3D microenvironments as they migrate through different tissue types. The migration mode that a cell adopts depends upon its microenvironment. Single cancer cells migrating in a mesenchymal mode are typically elongated or spindle shaped, with prominent actin stress fibers and actin-rich lamellipodia or filopodia at their leading edges. They migrate efficiently on 2D surfaces but are slower in 3D (23, 41). Mesenchymal migration is utilized by cells in tissues with a high substrate density and low degree of confinement (22, 25). Cells migrating in a mesenchymal mode release proteases to degrade the extracellular matrix and create new pathways for their movement, while using strong focal adhesions to latch on to the matrix and propel themselves forward (25). Increased Rac signaling is associated with this mesenchymal mode of migration (42–45).

Individual cancer cells may alternatively migrate in a distinctly different amoeboid mode. Amoeboid cells are characterized by a round morphology with membrane blebs that help drive locomotion (22, 23). Amoeboid cells lack actin stress fibers, have a nonfocal distribution of weak cell adhesion proteins, and exhibit high compressibility and deformability (20, 34, 37). These nonadhesive, round cells migrate efficiently in tissues with low matrix density with confinement (37), and their deformability allows them to navigate through tight spaces as they utilize existing pathways of least resistance (22). Increased activity of the RhoA-ROCK-myosin II signaling axis promotes actomyosin contractility, which drives this mode of amoeboid migration (35–39).

Individual cancer cells can transition between migration modes, allowing them to adapt to their changing microenvironment to optimize migration efficiency (23–25). Processes known as mesenchymal-amoeboid transition (MAT) or amoeboid-mesenchymal transition (AMT) enable cells to convert from one migration mode to another (23–25, 46–49). The nerve microenvironment appears to be conducive to an amoeboid form of migration since the nerve is a natural 3D tubular structure that serves as an existing pathway for cancer spread (22, 50). Nerves likely provide an environment of confinement within the intraneural or perineural space where space is limited. Deformable cancer cells in an amoeboid mode might migrate more efficiently within tubular neural structures as compared to a mesenchymal mode. Amoeboid migration does not utilize strong adhesive forces and enables rapid motility in 3D spaces with low matrix density using Rho-driven actomyosin contractility and membrane blebbing mechanisms. Confined tissue niches such as the intraneural space may induce a shift in cancer migration mode. Cancer cells possess innate plasticity enabling them to modulate their migration mode in response to the physical characteristics of their microenvironment such as matrix density, available space, and natural tissue barriers (25). This mechanism appears to be nonnerve specific and related to matrix density, available space, tissue barriers, and other microenvironmental factors as well as innate cancer cell plasticity. We have previously described multiple other nerve-specific mechanisms of cancer recruitment that may be occurring concurrently during PNI (9–14).

Cancer cells undergo shape changes in transitioning to an amoeboid migration phenotype and use their nuclei as mechanosensors of the environment (51, 52). Under confinement, the nuclear membrane stretches and unfolds, releasing calcium that activates PLA2, which catalyzes the formation of arachidonic acid, leading to the contractility of the myosin cortex, and allowing fast movement of amoeboid cells. This migration switch is a nongenetic program occurring in a short time period. In our model, cells have undergone transcriptomic changes. It is possible that in addition to these rapid changes that allow rapid migration in nerves, the nuclear membrane of the cells might be damaged and induce chromatin changes altering transcription (53).

Leading cells may be pharmacologically transitioned from amoeboid to a mesenchymal phenotype by the application of a ROCK inhibitor. Their phenotype becomes less round, more spindle shaped, and show increased expression of focal adhesions, increased filipodia, and decreased expression of blebs. Leading cells treated with ROCK inhibition also exhibit diminished ability to pass through constrictions in microchannels and diminished neurite association with dorsal root ganglia in gel matrix. Murine sciatic nerves invaded by neuroinvasive pancreatic cancers in vivo showed diminished PNI invasion length following ROCK inhibition, although limb function was not impacted. We speculate that nerve function might be more likely to be retained when deformable, migrating amoeboid cancer cells insinuate along sciatic nerve fibers and are less likely to disrupt axonal integrity as compared with protease-releasing, invading mesenchymal cancer cells.

Collectively, these findings highlight the plasticity of cancer cell migration mode in nerves and show that accelerated PNI is associated with cancer cell mesenchymal to amoeboid transition. Increased expression of ROCK1 and ROCK2 in amoeboid, neuroinvasive cells can be targeted pharmacologically to revert cells to a mesenchymal phenotype. This approach identifies cancer migration mode as a therapeutic target and reveals opportunities to treat this highly aggressive form of cancer invasion.

Materials and Methods

Cell Lines and Cell Culture.

The human pancreatic cancer cell line MiaPaCa-2 was purchased from ATCC. Cells underwent nucleofection to generate red fluorescent MiaPaCa-2 (MiaPaCa-2 RFP), using Kit V and program B-024 (Amaxa Nucleofector, Lonza Bioscience). Stable cell lines were generated by selection with G418 (100 µg/mL) (12). Cells with the highest expression level of RFP were selected by flow cytometry using a BD FACSAria from the Memorial Sloan Kettering Cancer Center (MSKCC) facility. Cells were cultured in 5% CO2 at 37 °C in DMEM (Cellgro) containing 10% fetal bovine serum (Gemini) and 50 U/mL penicillin/streptomycin (Cellgro). Culture media for cells expressing RFP were supplemented with G418.

MTT Assay.

Cancer cells (2,000) were plated in 96-well plates (Corning, 353910) overnight. For ROCK inhibitor treatments, cells were treated with Y27632 (Abcam, ab120129) at 5, 10, 20 μM and control (0.1% DMSO) in 100 μL/well the next day. MTT assay was performed at 24, 48, 72, 96, and 120 h after plating for proliferation experiments, or after inhibitor treatment. MTT solution was added at 10 μL/well (5 mg/mL, Sigma, M2128) after each time point was reached. After 4 h of incubation, 100 μL isopropanol (Fisher, BP2618) with 0.04N HCl (Sigma, 1090571000) was mixed in each well. The plate was read by The Synergy HT plate reader (Biotek) at 550 nm, reference 650 nm.

Sciatic Nerve Injections.

Athymic nude (NU/J) mice were obtained from The Jackson Laboratory (Bar Harbor, ME). Injection of MiaPaCa-2 RFP cancer cells into murine sciatic nerves was performed as previously described (18). Briefly, mice were anesthetized using 2% isoflurane, and their sciatic nerves were exposed. Concentrated cancer cells (2 µL, 150,000 cells) were injected into the sciatic nerve near the distal femur, with the needle pointed proximally toward the spinal cord using a 10-µL Hamilton syringe under magnification. After treatment with bupivacaine for analgesia, the incision was closed using 5–0 nylon sutures. Mice were examined for the first 3 postoperative days, and sciatic nerve function was measured weekly using a validated sciatic neurological score (hind limb motor response to extension, 4 = normal and 1 = full paralysis) (8, 17). Mice were killed when the majority had a sciatic neurological score of 2 or lower. All animal studies were conducted in accordance with an institutional protocol approved by the Memorial Sloan Kettering Institutional Animal Care and Use Committee.

Sciatic Nerve Extractions and Assessment of PNI.

Murine sciatic nerves were dissected from the distal femur to the spinal cord, extracted, and the length of cancer invasion along the nerve was measured from the distal edge of the primary tumor to the most proximal portion of the thickened nerve using a Vernier caliper. The most proximal 5 mm and most distal 5 mm of the invaded nerve were collected separately and immediately digested for FACS. Specimens were pooled for the analysis (mean n = 7 mice per passage). Mice were excluded if no primary tumor was identified after cancer cell injection, or if they expired prior to the experimental end point.

Sciatic Nerve Digestion and Flow Cytometry.

For FACS, the most proximal 5 mm and most distal 5 mm of the invaded nerve were separately washed in cold PBS, minced, and subsequently digested in an enzyme mix of Collagenase IV (20,000 units, Worthington) and 40 µL DNAse 1 at 37 °C for 40 min to obtain single-cell suspension. Cells expressing RFP were selected by flow cytometry using BD FACSAria from the MSKCC facility. RFP cells obtained from the most proximal 5 mm of the invaded nerve were termed Lead-1 cells and those from the most distal 5 mm of the invaded nerve were termed Lag-1 cells. Approximately half of both the leading and lagging cells obtained were regrown in culture, and the other half used for RNA extraction and sequencing.

Serial Passaging of Cancer Cells in Murine Sciatic Nerves.

RFP expressing Lead-1 cells isolated by FACS were grown in sterile culture. Once confluent, cells were again injected into the sciatic nerves of athymic nude mice as described above. The cycle of harvesting, growth in vitro, and sciatic nerve injection was repeated serially four times, creating five new cell lines with numbers representing the passage number: Lag-1, Lead-1, Lead-2, Lead-3, and Lead-4. Each cell line was maintained in culture at 37 °C, and early passage aliquots were frozen and maintained at −80 °C.

RNA Sequencing and Whole Exome Sequencing.

Immediately following FACS, half of the RFP-expressing cells sorted from FACS were pelleted, and RNA was extracted using the RNeasy Micro Kit (Qiagen #74004). RNA Sequencing was performed by the MSKCC core facility. After RiboGreen quantification and quality control by Agilent BioAnalyzer, 2 ng total RNA with RNA integrity numbers ranging from 9.5 to 10 underwent amplification using the SMART-Seq v4 Ultra Low Input RNA Kit (Clonetech catalog #63488), with 12 cycles of amplification. Subsequently, 10 ng amplified cDNA was used to prepare libraries with the KAPA Hyper Prep Kit (Kapa Biosystems KK8504) using eight cycles of PCR. Samples were barcoded and run on a HiSeq 4000 in a PE50 run, using the HiSeq 3000/4000 SBS Kit (Illumina). An average of 34 million paired reads was generated per sample, and the percent of mRNA bases per sample ranged from 63 to 83%. Sequence data processing was performed by the Bioinformatics Core at MSK. Output data (FASTQ files) were mapped to the human genome using the rnaStar aligner that mapped reads genomically and resolved reads across splice junctions. Output SAM files were postprocessed using PICARD tools and converted into BAM format. The expression count matrix from the mapped reads was computed using HTSeq (http://htseq.readthedocs.io/), and the raw count matrix was processed using the R/Bioconductor package DESeq (https://www.bioconductor.org/packages//2.10/bioc/html/DESeq.html) to normalize the full data set and analyze differential expression between sample groups. Data were analyzed using GO pathway enrichment analysis with a web-based tool (https://amp.pharm.mssm.edu/Enrichr/) for using the KEGG 2019 dataset. Heatmaps were generated using Morpheus.

Immunofluorescence Microscopy.

Cancer cells seeded on glass coverslips at a concentration of 50,000 cells/well in 24-well plates and grown at 37 °C for 1 to 2 d. Cells were then fixed in 4% PFA for 10 min and blocked in 0.1% Triton and 1% horse serum for 1 h. Primary antibodies diluted in 0.1% horse serum/PBS were incubated overnight at 4 °C. The following day, cells were incubated with secondary antibodies (anti-mouse Alexa Flour 488, anti-rabbit Alexa Fluor 488, 1:500, Invitrogen) for 1 h, washed, and mounted on glass slides (ProLong Diamond Antifade Mountant with DAPI, Invitrogen) and stained with DAPI. Images were acquired using (Zeiss Axiovert 200M) at 40× objective (0.75NA). Confocal microscopy was performed using an oil objective (63×, NA 1.4, HCX PL APO) with an SP5 Leica confocal microscope. Fluorescence was measured using ImageJ (v 1.52q), and analysis was performed using GraphPad Prism (v 8.4.1). Stress fiber quantification was performed by counting the number of intracellular phalloidin staining linear fibers. Velocity was based on quantifying migration distance traveled over time during time lapse and measured as µm per minute. We used ImageJ to quantify focal adhesion length and cell roundness by assessing cell area and cell perimeter length. Roundness was calculated using the formula ${\displaystyle \raisebox{1ex}{$\mathbf{(}\mathbf{4}\mathbf{\times }\mathbf{\pi }\mathbf{\times }\mathbf{a}\mathbf{r}\mathbf{e}\mathbf{a}\mathbf{)}$}\!\left/ \!\raisebox{-1ex}{${\mathbf{p}\mathbf{e}\mathbf{r}\mathbf{i}\mathbf{m}\mathbf{e}\mathbf{t}\mathbf{e}\mathbf{r}}^{\mathbf{2}}$}\right.}$ , where a value of 1.0 indicates a perfect circle and values approaching 0 indicate progressively elongated shapes.

Immunofluorescence microscopy was on human pancreatic ductal adenocarcinoma (PDAC) surgical specimens. As controls, nonneoplastic pancreatic tissue was used from pancreatic surgical specimens of patients with either benign serous cystadenoma or ampullary tubulovillous adenoma. Co-staining was performed for s100 to identify nerves. Paraffin slides underwent deparaffinization at 60 °C, rehydration, and heat-induced antigen retrieval at 95 °C for 20 min (Biocare, DC2012). Subsequent blocking and antibody incubations steps were the same as described above for IF microscopy.

The following primary antibodies were used at the indicated dilutions: PODXL 1:100 (Abcam, ab150358), EZR 1:200 (Abcam, ab4069), GSN 1 µg/mL (Abcam, ab74420), ROCK1 1:100 (Abcam, ab134181), ROCK2 1:100 (ab125025), Paxillin 1:100 (Abcam, ab32115), Phalloidin 1:1,000 (Abcam, ab176753), VCL 1:100 (Abcam, ab129002), s100 1:100 (Abcam, ab4066).

RhoA FRET Microscopy.

The pEGFP-RhoA biosensor plasmid (Addgene, 68026, deposited by Michael Glotzer) was obtained. Plasmids were grown in LB broth at 37 °C for 16 h and DNA extracted (QIAGEN, 12165). DNA concentration was measured was using a NanoDrop spectrophotometer (ThermoFisher, ND-ONEC-W). For transfection, 5 μg DNA was transfected to cancer cells (4 × 10⁶ cells) via electroporation using the Nucleofector system (Lonza, AAB-1001). Cells (250,000) were seeded on a 35-mm glass-bottomed dish. FRET imaging was taken 72 h after transfection using an inverted confocal microscope (Leica, SP5). For FRET calculation, Donor and FRET images were masked using the thresholded Donor image. FRET efficiency was calculated as sum intensity FRET/(sum intensity Donor + sum intensity FRET), then normalized to area of signal (per pixel) for each image. FRET ratio images were calculated by dividing the masked FRET image by the masked Donor image.

Western Blot.

Lag-1 and Lead-4 cells (50,000 cells) were plated in 24-well plates (Corning, 353047) overnight. Cells were rinsed with PBS and lysed in 50 μl Laemmli SDS buffer (ThermoFisher, J60660-AC) with sample reducing agent (ThermoFisher, NP0009) and denatured for 20 min at 95 °C. Samples were loaded to 4 to 15% gel (Bio-Rad, 456-1086), run at 100 V, and transferred onto a nitrocellulose membrane (Bio-Rad, 1704158) using Trans-Blot Turbo Transfer System (Bio-Rad). Membranes were blocked with blocking buffer (LI-COR, 927-70001) for 1 h and incubated with primary antibodies (Cell Signaling, cofilin 5175, 1:1,000; phospho-cofilin 3313, 1:1,000) overnight and 1 h for secondary antibodies (LI-COR, 926-68070, 926-68071, 926-32211, all 1:10,000), both prepared in blocking buffer with 0.2% Tween 20 per the manufacturer’s instructions. Images were taken with the Odyssey DLx imager (LI-COR) and analyzed with LI-COR image studio software (LI-COR).

Atomic Force Microscopy.

Glass-bottomed petri dishes (FluoroDish FD5040, World Precision Instruments) containing Lead and Lag cells were used. Experiments were performed at 37 °C with an MFP-3D-BIO AFM microscope (Oxford Instruments). Images and stiffness maps were obtained using the inverted optical microscope (Zeiss AxioObserver Z1) integrated with the AFM microscope. Cantilevers with colloidal borosilicate probes were used for experiments with a diameter of 5 μm and nominal spring constant k = 0.1 N/m (Novascan). Before each experiment, the exact cantilever spring constant was determined with the thermal noise method and the optical sensitivity was determined using a glass-bottomed petri dish filled with PBS as an infinitely stiff substrate. Stiffness maps of 80 × 80 μm² (20 × 20 points) were collected in areas containing the cells and the substrate, used as reference, at the rate of 1.5 Hz for a complete single approach/withdraw cycle. A trigger point of 1 nN was used to ensure maximum sample penetration of less than 1 micron. Force curves in each map were fitted according to the Hertz model using the routine implemented in the MFP3D AFM (Igor Pro, WaveMetrics). Data fitting was performed in the range from 0 to 60% of the maximum applied force, by setting tip Poisson = 0.19, tip Young’s modulus = 68 GPa, and sample Poisson = 0.45. To account for possible substrate effects, a threshold of 500 nm was chosen for the selection of the stiffness data, from the corresponding topographical maps of each cell collected in situ with the stiffness maps. Extraction of the stiffness values from the raw Igor Binary Wave data within the mask region was obtained by means of a home-built routine implemented in Igor Pro (WaveMetrics) and a custom script in MATLAB (MathWorks). Data were visualized using OriginPro (OriginLab) and GraphPad Prism.

2D Live Cell Imaging.

Cancer cells (30,000 cells) were plated in glass-bottomed culture dishes (FluoroDish FD5040, World Precision Instruments). Cells were incubated for 24 h before imaging. Cells were imaged by microscopy (Zeiss Axiovert 200M) equipped with 10× objective (0.3NA) lens, motorized stage, and temperature and CO2 controllers. Cells were kept at 37 °C with 5% CO2. Images were recorded every 10 min for 24 h. Single-cell tracking was performed in ImageJ using the “manual tracking” plugin, using their approximate centroid location over time. Only single cells were tracked, and tracking was stopped if the cell experienced a collision with another cell, underwent cell division or cell death, or migrated out of the field of view. Tracking figures, and distance and velocity calculations were performed using the Chemotaxis and Migration Tool (Ibidi) v2.0.

Microchannel Live Cell Imaging.

Microchannels were obtained from 4Dcell, 12-µm-wide and 10-µm high microchannels, of 150-µm length (MC005). Microchannels were used of the same dimensions with constrictions of 8 µm width and 5 µm length (MC11). Microchannel plates were prepared by coating with fibronectin (10 μg/ml) (Sigma-Aldrich F1141) at 37 °C for 1 h. Cancer cells (50,000 cells) were placed in microchannel wells and kept in the incubator overnight before imaging. Cells were imaged every 10 min for 24 h using the Zeiss Axiovert 200M with 10× objective (0.3NA) microscope at 37 °C and 5% CO2. Migration through the constrictions was analyzed using Axiovision.

ROCK Inhibition In Vitro and In Vivo.

Athymic nude (NU/J) mice were obtained from The Jackson Laboratory (Bar Harbor, ME). MiaPaCa-2 cancer cells were pretreated with the ROCK inhibitor Y27632 (25 µM, Abcam, ab120129) for 24 h prior to injection. When preparing cells for injection, they were bathed and centrifuged in 25 µM Y27632-treated media. Injection of MiaPaCa-2 cancer cells into murine sciatic nerves was performed as described above (control, n = 10; inhibitor, n = 10; male and female 50:50 mix). Mice were subsequently treated three times a week with 1 mg (female) to 1.25 mg (male) ROCK inhibitor Y27632 (5 mg/mL) via intraperitoneal injection. Mice were euthanized at 4 wk, and their sciatic nerves were dissected from the distal femur to the spinal cord. The length of cancer invasion along the nerve was measured from the proximal edge of the primary tumor to the most proximal portion of the thickened nerve using a Vernier caliper. Mice were excluded if no primary tumor was identified after cancer cell injection. Analysis was performed using GraphPad Prism (v 8.1.1).

Human Cancer Cell Line Expression of a 6 Gene Signature.

We assessed RNA sequencing data from a panel of over 1,400 cell lines using the CCLE from the Broad Institute. The expression was assessed of the 6 genes up-regulated in Lead cells identified in Fig. 2C (ROCK2, PXN, GSN, PODXL, VCL, and MYO10). The expression level of each gene was scaled considering 100% equal to the maximum expression level of any cell line in the CCLE database. A composite score of all 6 genes was then summed together to create an overall index. This methodology assumes an equal importance to each of the genes. The cell lines with the 24 highest and lowest sum scores were assessed.

Statistical Analysis.

Pairwise comparisons were conducted using an unpaired two-tailed Student’s t test for continuous data and Fisher’s exact test for categorical data. Statistical significance was defined as P ≤ 0.05. Statistical analyses were performed using Prism 8 (GraphPad Software, Inc.).

Supplementary Material

Appendix 01 (PDF)

Movie S1.

Lead cell in a 3D microchannel showing amoeboid migration.

Movie S2.

Lag cell in a 3D microchannel showing mesenchymal migration.

Movie S3.

Lead cell passes through a constriction in a 3D microchannel.

Movie S4.

Lag cell at a constriction in a 3D microchannel.

Data, Materials, and Software Availability

RNA sequencing data are publicly available at the GEO repository under accession number GSE207319 (54).