Homophilic CD44 Interactions Mediate Tumor Cell Aggregation and Polyclonal Metastasis in Patient-Derived Breast Cancer Models

Abstract

Circulating tumor cells (CTC) seed cancer metastases; however, the underlying cellular and molecular mechanisms remain unclear. CTC clusters were less frequently detected but more metastatic than single CTCs of patients with triple-negative breast cancer and representative patient-derived xenograft models. Using intravital multiphoton microscopic imaging, we found that clustered tumor cells in migration and circulation resulted from aggregation of individual tumor cells rather than collective migration and cohesive shedding. Aggregated tumor cells exhibited enriched expression of the breast cancer stem cell marker CD44 and promoted tumorigenesis and polyclonal metastasis. Depletion of CD44 effectively prevented tumor cell aggregation and decreased PAK2 levels. The intercellular CD44-CD44 homophilic interactions directed multicellular aggregation, requiring its N-terminal domain, and initiated CD44-PAK2 interactions for further activation of FAK signaling. Our studies highlight that CD44⁺ CTC clusters, whose presence is correlated with a poor prognosis of patients with breast cancer, can serve as novel therapeutic targets of polyclonal metastasis. SIGNIFICANCE: CTCs not only serve as important biomarkers for liquid biopsies, but also mediate devastating metastases. CD44 homophilic interactions and subsequent CD44-PAK2 interactions mediate tumor cluster aggregation. This will lead to innovative biomarker applications to predict prognosis, facilitate development of new targeting strategies to block polyclonal metastasis, and improve clinical outcomes.See related commentary by Rodrigues and Vanharanta, p. 22.This article is highlighted in the In This Issue feature, p. 1.

Figure 1.. Tumor cell clusters arise from cellular aggregation

A. H & E staining images of CTC clusters (orange arrows) within the vasculature of the lung metastasis sections of TNBC patient CW1 (left panel) and a TN1 PDX mouse (right panel). Scale bars=10 μm. B. IHC staining with a TN PDX breast tumor section for cytokeratin (CK) showing clustered tumor cells within the vasculature (a lower magnification image is in Supplementary Fig. S1B). Scale bar=10 μm. C. Frequencies of IHC-detected vascular CTC clusters (% of all CTC events) within breast tumor and distant metastasis sections of seven patients (n=9 human tissues) and seven PDX models (n=28 mouse tissues) (listed in Supplementary Table S1). T-test p=0.1115 (NS). D. Human CTC clusters in the peripheral blood of patients with metastatic breast cancer, negative for CD45 and positive for pan-CK and nuclear DNA (DAPI), detected via EpCAM-based CellSearch platform. Scale bars=10 μm. E. Fluorescence images of TN1 PDX tumor cell clusters within the peripheral blood and the lungs of NOD/SCID mice. Top panel: blood CTC cluster (tdTomato⁺) from L2T PDX-bearing mice (blue: Hoechst). Bottom panel: 3D stack image of a dual-color lung colony with one L2G (eGFP⁺) cell and one L2T (tdTomato⁺) cell derived from mixed-color implants as shown in Supplementary Fig. S2A. Scale bars=10 μm. F. Frequencies of blood CTC clusters (% of all CTC events) isolated from seven patients with metastatic breast cancer (n=7) and mice with four PDX models (n=7 mice) (Supplementary Table S2). T-test p=0.533 (NS). G. Intravital images of TN1 PDX breast tumor cell cluster formation via cell aggregation during migration, showing individually migrating eGFP⁺ tumor cells approaching and aggregating with other tumor cells and moving around dynamically. Arrows at 24’ and 30’ show the cumulative paths of cells 1, 2, and 3. Red = dextran+ vessels, blue = second harmonic generation (collagen I fibers). Scale bar=10 μm. See Supplementary Video S1. H. Intravital images of single-cell intravasation of eGFP⁺ MDA-MB-231 tumor cells following cluster formation in a primary tumor. Stationary tumor cell 1 is joined by individually migrating cells 2 & 3 to form a cluster. Cell number 2 sequentially leaves the cluster and intravasates between the frames at 18’ and 20’. Green = tumor cells, red = vasculature. Scale bar=10 μm. See Supplementary Video S3. I. Intravital images of eGFP⁺ PyMT breast tumor cells (yellow in the red vasculature) in MacBlue Rag^−/− mice, circulating as single cells (T1 and T5) and as groups of cells (T2, T3, and T4) in close physical proximity to each other. Tumor cells are briefly observed as they rapidly pass through the imaging field due to blood flow. Green = tumor cells (CTCs shown yellow), red = 155kD TMR–dextran-labeled vasculature, cyan = macrophages (circulating monocytes in white in the red vasculature). Scale bar = 10 μm. Additional CTCs in Supplementary Video S4. J. Patient-derived CTC line BRX50 cells form clusters within one to two hours of suspension culture. Scale bar=50 μm. K. Cluster formation within the lung vasculature imaged ex vivo at 2 h after tail vein infusion of eGFP⁺ (green) and tdTomato⁺ (red) MDA-MB-231 cells at 1:1 ratio, either mixed co-infusion (0 min apart), or separate infusions of tdTomato⁺ cells first and then eGFP⁺ cells lagged at 5 min, 10 min, and 2 h. Ex vivo lung fluorescence images were taken 2 h post-infusion of eGFP⁺ cells. Scale bars=50 μm. L. Quantitative proportions of single-color and mixed-color clusters (lung colonies) from the four groups in panel K. The experiments were repeated three times (n=3) with counts of at least five images per mouse. T-test, ***p<0.001.

Figure 2.. Tumor cell clusters with increased tumorigenesis, metastasis, and CD44 expression

A. Representative bioluminescence images of single cells (S) and clusters (C) of TN1 PDX tumor cells in parallel during tumorigenic monitoring upon orthotopic implantation (1,000 cells per mammary fat pad), on days 0 (D0) and 18 (D18). B. Quantitative bioluminescence signals (total flux, p/s) (left panel) and fold change (right panel) of tumorigenesis mediated by TN1 tumor cells in singles and clusters during the 18-day monitoring in panel A (n=5). T-test **p<0.01. C. Representative bioluminescence images of single cells (S) and clusters (C) of TN2 PDX tumor cells in parallel during tumorigenic monitoring upon orthotopic implantation (5,000 cells per mammary fat pad), on day 0 (D0) and day 24 (D24). D. Quantitative bioluminescence signals (total flux, p/s) (left panel) and fold change (right panel) of tumorigenesis mediated by TN2 tumor cells in singles and clusters during the 24-day monitoring in C (n=8). T-test *p<0.05, **p<0.01, ***p<0.001. E. Images of mammospheres derived from single and clustered tumor cells of TN2 PDXs. Scale bars=50 μm. F. Quantitated bar graph of mammospheres derived from single cells and clusters. N= 6 biological replicates. ***p=0.0008. G. BLI images of lung colonization mediated by single cells and clusters of L2T-labeled TN1 PDX cells (5×10⁵) at day 0 (D0) and weeks 1, 2 and 8 (Wk1, 2, 8) after tail vein infusion. H. Quantitated BLI signal (total flux, p/s) of the TN1 single cell- and cluster-mediated lung metastases described in panel G (n=5). T-test *p=0.012, ***p=0.00012. I. CD44 (brown) and CD31 (pink) IHC staining of the TN1 PDX-bearing mouse lung sections (slide 228) showing a CD44⁺ CTC cluster within CD31⁺ vasculature. Scale bar=25 μm. J. Proportion of CD44⁺ CTCs in the events of single and cluster CTCs counted within the in situ vasculature of human breast tumors (n=5) and metastases (n=3), and PDX lung metastases (n=3) (Supplementary Table S3). T-test ****p<0.0001.

Figure 3.. CD44 directs polyclonal tumor cell aggregation

A. Time lapse aggregation images at 0, 4, 8, and 24 h incubation with PDX-derived 1:1 mixtures of tdTomato+ and eGFP+ primary tumor cells. Left column: sorted CD44⁺; middle column, CD44^-; right column, mixed CD44⁺/CD44^- cells in two colors. For details see the Supplementary Videos S6-S8. B. Number of viable CD44⁺ versus CD44^- cells at the time points of 12 and 24 h aggregation (T-test p=0.6, n=4). C. Immunoblot of CD44 and β-actin (loading control) in TN1 PDX tumor cells upon transfection of the scrambled siRNA control (siCon) or siCD44, which caused a depletion of the dominant variant CD44 (CD44v, molecular weight 120~160 kDa) and the marginal standard CD44 (CD44s, molecular weight ~80 kDa). D. CD44 immunoblot showing siCD44-mediated knockdown of the exclusive CD44s in MDA-MB-231 cells compared to the control (siCon). E. CD44 immunoblot of three MDA-MB-231 batches of populations upon CRISPR/Cas9-based CD44 knockout (KO) with gRNA targeting exon 2. F. Snapshot images showing reduced aggregation of L2G-labeled TN1 PDX tumor cells at 72-h clustering time point after siCD44 and siCon transfections, measured via IncuCyte imaging system. G. Quantitative curves of the cluster size (left panel) and number (right panel) of TN1 cells measured via IncuCyte time-lapse imaging (n=3 biological replicates, MANOVA ***p<0.001). H. Snapshot images taken at 0 and 60 min of MDA-MB-231 suspension cell aggregation onto poly-hydroxyethyl methacrylate-treated plates, at 48 h after siCD44 and siCon transfections. Scale bars=50 μm. I. Cluster counts of MDA-MB-231 cells at 60 min aggregation. The data were from at least five images for each group per experiment. The experiments were repeated three times (n=3). T-test *p<0.05, **p<0.01, ***p<0.001. J. Cluster images of CD44 WT and KO (via CRISPR/Cas9) MDA-MB-231 cells at the 24 h aggregation time-point. Scale bars=150 μm. K. Quantitative counts of clustered MDA-MB-231 cells with a cluster size >20 cells show a significant, dramatic reduction in CD44 KO cells (n=5, T-test ***p<0.001).

Figure 4.. CD44 depletion blocks tumor cell aggregation and lung metastasis in vivo

A. Bioluminescence images of lung colonization of the siRNA control (siCon) and siCD44-transfected TN1 tumor cells on days 0 (D0) and 1 (D1) and weeks 1 (Wk1) and 4 (Wk4) post-tail vein infusion. 5×10⁵ cells were injected into mice at 36 h after the initial transfection. B. Quantitative bioluminescence signal curves (total flux, p/s) of the TN1 PDXs (siCon and siCD44) as measured in A, n=5 mice per group. T-test ***p<0.001. C. Bioluminescence images of lung colonization of the siRNA control (siCon) and siCD44-transfected TN2 tumor cells on days 0 (D0) and 1 (D1) and week 1 (Wk1) post-tail vein infusion. 5×10⁵ cells were injected to mice at 36 h after the initial transfection. D. Quantitative bioluminescence signal curves (total flux, p/s) of the TN2 PDXs (siCon and siCD44) as measured in C, n=5 mice per group. T-test **p<0.01. E. Bioluminescence images of lung colonization of L2T-labeled CD44 WT and KO MDA-MB-231 tumor cells on day 0 (D0) and weeks 1 (Wk1), 3 (Wk3), and 5 (Wk5) post-tail vein infusion. F. Quantitative analyses of the lung bioluminescence signals of CD44 WT and KO cells imaged in E (n=4). T-test *p<0.05, ****p<0.0001. G. Fluorescence images of the lungs at 2 and 24 h and 5 weeks post-tail vein infusion of mixed eGFP⁺CD44KO and tdTomato⁺CD44WT tumor cells (1:1 ratio). Three columns represent the images from the tdTomato channel, the eGFP channel, or the merged channels, respectively. Scale bars=50 μm for the images taken at 2 and 24 h, and 125 μm for the images taken at week 5. H. Counts of single (solid bar) or clustered tumor cells (checked bar) of CD44 WT and KO cells in the lung images at 2 and 24 h post-tail vein injections in E. At least five lung images were taken for each mouse (n=3 mice). I. Tumorigenesis results of PDX-derived CD44⁺ WT and CD44^- KO cells implanted into the mammary fat pads (2^nd and 4^th), from 1,000 to 100,000 cells per injection. *T-test p=0.02 between WT/KO implantations of 1,000 cells. J. Top: Depiction of orthotopic implantation of CD44 WT (eGFP⁺) and KO (tdTomato⁺) tumor cells (100,000 cell per injection) separately into the left and right mammary fat pads of each NOD/SCID mouse. Bottom: images of breast tumors derived from the above implantations at harvest (3 weeks). K. Comparison of tumor weight of CD44 WT (eGFP⁺) and KO (tdTomato⁺) tumors derived from 100,000 cell injections in J (n=6 mice). T-test ***p<0.001. L. Fluorescence images, from the channels of tdTomato (left), eGFP (middle), and merged (right), of the dissected lungs with spontaneous metastases of CD44 WT (eGFP⁺) and KO (tdTomato⁺) tumors, at 3 weeks post-orthotopic implantation of these cells into the left and right 4^th mammary fat pads, respectively (as shown in J). Scale bars =125μm. M&N. Comparison of lung colonies (M) and normalized lung metastases (colony # per gram of tumor weight) (N) following implantations of CD44 WT and KO cells. T-test ****p<0.0001.

Figure 5.. CD44 mediates cell aggregation via intercellular, homophilic interactions

A. Aggregation images at the 72-h time point showing that hyaluronan antagonist (o-HA) slightly increased the cluster size of L2T-labeled TN1 PDX tumor cells. Scale bars=50 μm. B. Quantitative curves of the cluster size (left panel) and number (right panel) of TN1 cells measured via IncuCyte time-lapse imaging (n=3 biological replicates, MANOVA *p<0.05). C. Images of MDA-MB-231 cell aggregation at 1 h with or without hyaluronic acid synthase inhibitor (HASi) 4-MU (at 0.4 mM/L) pre-treatment for 48 h. Scale bars=25 μm. D. Quantitative counts of aggregated MDA-MB-231 cells with cluster sizes of 2–5, 6–10 and >10 cells pretreated with or without 4-MU (n=5). NS=no significant difference. E. CD44 immunofluorescence staining with dissociated MDA-MB-231 cells during the aggregation assay 48 h post-transfection with siCon (top panel) and siCD44 (bottom panels). Most of the CD44-negative cells remained as single cells in suspension, whereas residual CD44 in knockdown cells was located at the intercellular interface of a few clusters. F. The binding curves of biotin-conjugated CD44 at 0, 1 and 5 μg/ml to the solid phase CD44 and BSA, measured as OD₄₅₀ units. T-test ***p<0.001. G. Top panel: diagram of mixed HEK-293 cell aggregates of two populations transfected with C-terminal FLAG-tagged and HA-tagged CD44, respectively. Bottom panels: immunoblots for the CD44-FLAG and CD44-HA proteins upon co-IP with anti-HA and anti-FLAG antibodies, respectively. H. Structure model of CD44s monomer (the signal peptide 1–20 not shown) with the N-terminal residues, especially Q21-C97 of the extracellular domain I and the beginning of the domain II, as predicted by computational algorithm iTasser. Warmer colors indicate higher probabilities to be at the dimer interface, as predicted by protein docking algorithm BAL. I. Two representative structure models (top and bottom panels) of predicted CD44 homodimers (formed between two neighboring cells) at an almost straight angle from protein docking. The right monomer is colored coded in the same way as in H whereas the left one is in gray for contrast (see supplementary Figure S8 for additional structure models). J. Immunoblots for CD44-FLAG (CD44s and ΔN21–97) and CD44-HA proteins upon co-IP using anti-HA and anti-FLAG antibodies with mixed HEK-293 cell aggregates of two populations transfected with FLAG-tagged (CD44s and ΔN21–97) and HA-tagged CD44, respectively. *CD44s wildtype or mutant bands. K. Images of aggregation of HEK-293 cells for 1 h, at 48 h post transfection with CD44s-FLAG and ΔN21–97-FLAG. Scale bars=50 μm. L. Quantitative counts of aggregated HEK-293 cells, transfected with CD44s-FLAG and ΔN21–97-FLAG, in cluster sizes of 2–5, 6–10 and >10 cells. T-test ***p<0.001.

Figure 6.. CD44 promotes PAK2 pathway in tumor cell aggregates

A. The number of proteins with a >2-fold change in CD44⁺ versus CD44^- and siCD44 versus control comparisons: 535 out of 1377, and 382 out of 1523, respectively, with 38 proteins in common. The graph shows the canonical pathways of the 38 overlapped proteins. B. Immunoblots of PAK2 in TN1 PDX tumor cells transfected with the control siCon, siPAK2, and siCD44, at 36 h after knockdown. Loading control: β-actin. C. Relative similar mRNA levels of PAK1 and PAK2 (NS=no significant change) upon siCD44 knockdown (***p<0.001), measured via quantitative real-time PCR. D. Representative aggregate images of tdTomato⁺ TN1 PDX tumor cells at 72 h aggregation upon PAK2 knockdown via siPAK2. Scale bars=50 μm. E. Quantitative analyses of cluster size (left panel) and number (right panel) of TN PDX tumor clusters upon siPAK2 knockdown, measured by IncuCyte time lapse imaging. MANOVA ***p <0.001. F. Bioluminescence images of lung colonization of the siCon, siPAK2, and siCD44-transfected TN PDX tumor cells on days 0 (D0) and 2 (D2) post-tail vein infusion (36-h post transfection). G. Quantitative bioluminescence signal curves (% of D0 signal) of reduced lung colonization of TN PDX cells upon knockdown via siPAK2 and siCD44 (n=5 mice per group). T-test *p<0.05 for both siPAK2 and siCD44 comparisons to the control siCon at both D1 and D2. H. Immunoblots for the tagged proteins PAK2-FLAG and CD44-HA upon co-IP with anti-HA (CD44) using the lysates of 293T cell aggregates, 48 h post-cotransfection with PAK2-FLAG and CD44-HA. I. IF staining images of endogenous CD44 and PAK2, and Dapi signals showing the high expression of co-localized CD44 and PAK2 at the cytoplasmic membrane of the aggregated MDA-MB-231 cells (24 h aggregation). In contrast, the single cell (white arrows) in suspension display low levels of CD44 and PAK2 which are not co-localized. Scale bar=20 μm. J. Immunoblots with MDA-MB-231 cell lysates for CD44, p-PAK2, total PAK2, p-FAK, and FAK detection at 48 h post-transfection with siCD44 and siPAK2. CD44 and PAK2 positively promote each other’s protein levels and FAK phosphorylation.

Figure 7.. CD44⁺ CTC cluster association with clinical outcomes

A-C. Kaplan-Meier plot of OS (A), RFS (B), and DMFS (C) for patients with high (red) and low (blue) CD44 (probes 31615_i or 210916_s) expression in breast tumors. Expression was dichotomized at the optimal cut point plotted in GSE3143 (n=158, left panel) and GSE7390 (n=198, right two panels). 95% confidence intervals for each group are indicated by dotted lines. Cox p=0.006, 0.026 and 0.008 as indicated. D. Kaplan-Meier plot of DMFS by PAK2 mRNA expression. High (red) and low (blue) groups were determined using the optimal cut point in the GSE19615 human breast tumor database (n=200). Cox p=0.014. E. Kaplan-Meier plot of OS for Northwestern breast cancer patients with cluster-positive and -negative CTCs, detected by CellSearch (n=118, Log rank test p=0.0057) (Supplementary Table S6). F. Representative images of a CD44⁺ cell cluster (top panels) and a single CD44^- CTC (bottom panels) detected in human peripheral blood via CellSearch platform staining for CD45, CD44, CK, and DAPI. Scale bar=10 μm. G. Bar graph of OS, based on the swimmer plot principle, for Northwestern breast cancer patients with CellSearch-detected CD44⁺ CTC clusters and CD44^- CTCs only (n=8, Log rank test p=0.0389). Two patients with CD44⁺ clusters were deceased due to disease progression (Supplementary Table S7). H. Diagram of individual cell migration leading to perivascular cluster formation and intravasation. Tumor cells individually migrate to sites of intravasation on blood vessels, where CD44 mediates intercellular, homophilic protein complex formation and subsequent CD44-PAK2 interaction and FAK pathway activation. The self-interaction drives tumor cell aggregation at and within the vasculature in the primary tumor and lung metastases, due to their physical proximity. CD44 directs aggregation of detached breast tumor cells that mediate polyclonal metastasis, while single CD44- tumor cells undergo anoikis within 48 to 72 h of detachment.