Targeting of Adenovirus Vectors to Tumor Cells Does Not Enable Efficient Transduction of Breast Cancer Metastases¹ Free

Patients (>70%) with metastatic cancer develop liver metastases (1). The majority of liver metastases is resistant to radio- or chemo-therapy. In these cases, gene therapy approaches are considered to be a potential alternative for tumor treatment (for review, see Ref. 2). The efficacy of strategies that involve transfer of growth inhibitory, proapoptotic, or tumor-sensitizing genes critically depends on the number of initially transduced tumor cells. Overall, transduction of tumor cells after direct intratumor injection is not efficient and is restricted to the injection site (3). Therefore, intravascular application of antitumor vectors is a prerequisite to reach all metastases and a maximal number of tumor cells within tumors. Ad³ vectors are attractive vehicles for in vivo gene transfer. Most Ad vectors are based on serotypes 2 or 5. Infections with these serotypes involve high affinity interactions between the viral fiber and a cellular receptor that was identified recently as the CAR (4). Although Ads efficiently infect a large variety of human tumor cell lines, transduction of primary tumors in patients or tumors established in animal models is often inefficient. Potential factors that limit the transduction of solid tumors include: (a) the level of CAR expression; it appears that CAR expression inversely correlates with the malignant potential of tumors, resulting in low infectivity of highly aggressive tumors (5, 6, 7, 8); (b) sequestration of systemically applied Ads in normal tissue; in tumor animal models, i.v. application often results in transduction of hepatocytes with little or no transduction of tumor cells (2, 9), and the preferential transduction of hepatocytes is associated with toxic side effects from expressed viral proteins or cytotoxic transgene products (10, 11, 12); and (c) anatomical barriers that limit the accessibility of tumors to Ad vectors. These include the tumor structure, vascularization (13), and ECM, which often surrounds liver metastases, particularly metastases derived from breast cancer (14). The ECM consists of the interstitium containing blood vessel, matrix proteins (collagens I, III, IV, VI, laminin, and fibronectin), and the basement membrane that forms a network throughout the ECM. The matrix proteins are synthesized and assembled by synergistic involvement of both epithelial tumor cells and adjacent stromal cells (15). The ECM has functional importance in tumor cell adhesion, migration, signaling, and angiogenesis (16). It promotes tumor survival by protecting against apoptosis (17) and inhibiting infiltration of and recognition by immune cells.

Many strategies have been pursued to enhance in vivo transduction with Ad vectors, including coinjection of polycations (18), organic solvents (19), or pretreatment with proteases (3, 20). However, these strategies are associated with toxicity and are not practicable in clinical settings. A number of approaches has been directed toward eliminating the promiscuous binding properties of Ad5-based vectors and providing a new selective ligand to the virus. Although, cost-intensive, retargeting strategies, including chimeric fusion proteins and bispecific antibodies, have yielded promising results (21, 22). Recently, genetic modification of the viral capsid has been used to directly introduce new targeting ligands (6, 23).

In the present study, we developed a mouse model of metastatic breast cancer to test whether the tropism of a chimeric Ad that recognized a receptor different from CAR could be exploited for specific targeting of liver metastases. We found that the transduction of metastases in vivo did not solely depend on the presence of high affinity receptors but was strongly influenced by the ECM surrounding metastatic nests and the accessibility of tumor cells to blood vessels.

Human embryonic kidney (293) cells, (Microbix, Toronto, Canada) were maintained in DMEM, 10% FCS, 2 mm glutamine, and Pen/Strep. Human breast cancer, MDA-MB435 cells (ATCC HTB-129) were maintained in L-15 medium supplemented with 15% FCS, 2 mm glutamine, Pen/Strep, and 10 μg/ml bovine insulin. Primary mouse hepatocytes were isolated by collagenase perfusion (24) and cultured in PRIMARIA dishes (Corning) in William’s E medium, supplemented with 10% FCS, 2 mm glutamine, and Pen/Strep. To generate pAdE1-RSVβ-gal, the β-galactosidase gene and the SV40 polyadenylation signal were transferred from pSV-β-galactosidase (Promega Corp., Madison, WI) into pΔE1sp1A-RSV (25) under the control of the RSV promoter. To generate pAdE1-MSCV-GFP, fragments containing the eGFP expression unit (pEGFP-1; Clontech, Palo Alto, CA) under the control of MSCV promoter (derived from pMSCVneoEB) were cloned into pΔE1sp1A (Microbix). Recombinant viruses were generated in 293 cells after cotransfection of pAdE1-RSVβ-gal or pAdE1-MSCV-GFP with pBHG10 (for Ad5) or pAdΔΨF35 (for Ad5/35), as described earlier (26). pAdΔΨF35 is based on pBHG-10 (Microbix), where the Ad5 fiber gene was substituted with the Ad35 fiber gene (27). The corresponding viruses were named Ad5.RSVbGal, Ad5.MSCVGFP, Ad5/35.RSVbGal, and Ad5/35 MSCVGFP. The viruses were propagated in 293 cells, purified, and titered for genomes and pfu as described elsewhere (27).

All experimental procedures involving animals were conducted in accordance with the institutional guidelines set forth by the University of Washington. Immunodeficient C.B-17/lcrCrl-scid-bg/BR (CB17) mice (Charles River, Wilmington, MA) were housed in specific pathogen-free facilities. To establish mouse models with liver metastases, animals were infused with 2 × 10⁶ MDA-MB435 cells through a permanently placed portal vein catheter (25). For histological analysis, 3 weeks after transplantation, livers were recovered and fixed in OCT compound or 10% formalin for cryosections or paraffin sections, respectively.

For attachment, internalization, and cross-competition studies, Ad vectors were metabolically labeled with [³H]methyl-thymidine, as described elsewhere (27). For attachment studies, 3.5 × 10⁵ cells were incubated for 1 h on ice with [³H]-labeled Ad at an MOI of 8000 genomes/cell in 100 μl of ice-cold adhesion buffer (DMEM supplemented with 2 mm MgCl₂, 1% BSA, and 20 mm HEPES). Then, the cells were pelleted by centrifugation for 4 min at 1000 × g and washed two times with 0.5 ml of ice-cold PBS. After the last wash, the cells were pelleted at 1500 × g, the supernatant was removed, and the cpm associated with the cell pellet was determined by a scintillation counter. To determine the number of internalized [³H]-labeled adenoviral particles, cells were incubated on ice for 1 h with the corresponding virus, washed with PBS as described above, resuspended in 100 μl of adhesion buffer, and then incubated at 37°C for 30 min. To remove noninternalized [³H]-labeled particles, cells were diluted 3-fold with cold 0.05% trypsin-0.5 mm EDTA solution, incubated at 37°C for an additional 5–10 min and pelleted. For cross-competition studies, 10⁵ cells were preincubated on ice with 100-fold excess of unlabeled competitor virus. Then, the thymidine labeled virus was added at an MOI of 8000 genomes/cell, and attachment was analyzed as described above.

MDA-MB-435 cells (2.5 × 10⁵) or freshly purified primary mouse hepatocytes were seeded per well (12-well plates) 1 day before infection. The next day, the number of attached cells per well was counted, and virus was added at the MOI indicated in the figure legends in 400 μl of growth medium. Cells were incubated for 6 h at 37°C. Then, the virus-containing media were removed, and the cells were washed once with PBS and incubated in normal medium for 24 or 48 h when the analyses of GFP or β-galactosidase reporter gene expression were performed.

For in vivo application, 10¹¹ Ad genomes/particles in 200 μl of PBS were injected by tail-vein infusion into mice with or without pre-established liver metastases. For in vivo transduction studies, mice were sacrificed 72-h postinfusion, and livers were processed for histological analyses.

Isolation of cellular DNA from mouse liver and Southern analysis was performed as described elsewhere (27). The ³²P-labeled DNA probes used for hybridization are described in the figure legends.

For detection of collagen, paraffin sections were stained with Masson Trichrome. For detection of β-galactosidase, laminin, and CD31 by immunohistochemistry, serial sections (5-μm thick) of liver tissue were fixed with 4% paraformaldehyde or cold acetone for 10 min before incubation with primary antibodies. For detection of β-galactosidase protein, an anti-β-galactosidase mouse monoclonal antibody (1:200; Promega Corp.) was used on paraformaldehyde-fixed sections. Specific antibody binding was detected with a secondary antibody labeled with Alexa Flour 488 (green; Molecular Probes, Inc., Eugene, OR). Blood vessels were stained with monoclonal rat anti-CD31 (platelet/endothelial cell adhesion molecule 1, 1:50; PharMingen, San Diego, CA) antibody on acetone-fixed sections in combination with a secondary antirat antibody, conjugated with Alexa Flour 488. The basement membrane component laminin was analyzed with polyclonal rabbit antilaminin antibody (1:500; Research Diagnostic, Inc., Flanders, NJ), followed by secondary antibodies, conjugated with Alexa Flour 568 (red). Cell nuclei were counterstained with 1 μg/ml 4′,6-diamidino-2-phenylindole (Sigma Chemical Co., St. Louis, MO). Staining for β-galactosidase enzyme activity on liver sections or in cultured cells in situ was performed with X-Gal as described earlier (25).

All statistical analyses were performed with standard MS-Excel software.

Recently, we have developed a capsid-modified Ad vector containing sequences encoding the Ad type 35 fiber shaft and knob instead of the Ad5 fiber gene (27). The chimeric Ad5/35 vector interacts with a cellular receptor different from CAR and infects human hematopoietic cells, which are relatively refractory to infection with standard Ad5-based vectors (26, 27). Here, we tested whether the new tropism of Ad5/35 could be exploited to achieve tumor-specific infection of liver metastases. As a tumor cell model, we selected the human breast cancer cell line MDA-MB 435, which has been used recently in mouse models with tumors transplanted into the mammary fat pad (28). Because our ultimate goal was to specifically target liver metastases in a mouse model, we performed in vitro tropism studies on MDA-MB435 cells and on primary hepatocytes isolated by collagenase perfusion of mouse liver. Flow cytometry demonstrated that MDA-MB435 cells were strongly positive for α_v integrins but expressed only very low levels of the primary attachment receptor for Ad5, CAR (Fig. 1). It has been demonstrated previously that mouse liver expressed high levels of CAR mRNA (29) and relatively low levels of α_vβ_3/5 integrins; the susceptibility of mouse hepatocytes to Ad5 infection did not depend on the level of α_v integrin expression (30).

Attachment and internalization with metabolically labeled virions demonstrated that Ad5 particles attached poorly to and were internalized inefficiently by MDA-MB435 cells. In contrast, Ad5/35 interacted very efficiently with MDA-MB435 cells (Fig. 2,A). Inversely, Ad5 attached efficiently to and was internalized by primary mouse hepatocytes, whereas the interaction of Ad5/35 with these cells was minimal. The tumor specificity, expressed as the ratio of viral particles attached to MDA-MB435 cells to particles attached to primary mouse hepatocytes, was 53 for Ad5/35 and 0.2 for Ad5. Accordingly, the specific binding of Ad5/35 to MDA-MB435 cells was ∼250 times greater than Ad5. Cross-competition studies between Ad5 and Ad5/35 viruses for binding to primary hepatocytes or MDA-MB435 cells agreed with previous observations that these two viruses interact with two different receptors (Fig. 2,B; Ref. 27). Transduction studies based on transgene expression (Fig. 2 C) corroborated the attachment data and demonstrated that: (a) primary mouse hepatocytes were refractory to infection with Ad5/35, whereas this vector efficiently infected MDA-MB435 cells; and (b) MDA-MB435 cells were refractory to Ad5 infection because of the lack of CAR; however, Ad5 infected primary mouse hepatocytes efficiently. These findings represent the basis for our in vivo tropism studies in animals with liver metastases derived from MDA-MB435 cells.

To study the in vivo liver tropism of Ad5 and Ad5/35, we first analyzed viral DNA concentration in mouse liver 72 h after tail-vein injection into naïve mice. Quantitative Southern blot analysis demonstrated that the amount of Ad5/35 genomes in mouse livers was ∼10 times less than that of Ad5 (Fig. 3,A). In agreement with this, staining for β-galactosidase activity on liver sections from mice that received 5 × 10⁹ pfu of Ad5 or Ad5/35 showed that, on representative sections, 95 +/− 5% and 7 +/− 3% of hepatocytes were transduced with the Ad5 and Ad5/35 vector, respectively (Fig. 3 B). Therefore, Ad5/35 demonstrated a reduced tropism for hepatocytes in vivo.

In humans, invasive breast carcinoma cells disseminate through the portal vein into the liver, produce angiogenic factors, and form vascularized metastases surrounded by ECM (31, 32). To create a clinically relevant model, we transplanted MDA-MB435 cells into the portal vein of mice. Multiple macroscopically and microscopically distinguishable tumor foci developed in the liver 2–3 weeks post-transplantation. Histological analyses of liver sections demonstrated that metastases are lined by basement membrane-like structures composed of collagen and laminin (Fig. 4,A, left panels). The presence of blood vessels in hepatic tumors was detected by an antibody specific to platelet/endothelial cell adhesion molecule 1 (CD31), a marker associated with endothelial cells (Fig. 4,A, top row, middle panel). CD31 staining, often overlapping with laminin staining, was confined to the tumor stroma and was not present within the tumor nests (Fig. 4,A, top row, right panel and bottom row, middle and right panels). This suggested that tumors were vascularized with blood vessels present in-between the tumor foci. Furthermore, in the liver parenchyma, there was a gradient of CD31 staining with the highest density and intensity around tumors, indicating neoangiogenesis stimulated by tumor and/or stromal cells (Fig. 4 A, middle panels). Thus, the morphological and histological appearance of MDA-MB435-derived hepatic metastases in our mouse model resembles that seen in humans.

For in vivo targeting studies, Ad5- or Ad5/35-based vectors at a dose of 5 × 10⁹ pfu/mouse were administered systemically by tail-vein infusion into mice with established liver metastases. At least six sections/mouse with five mice/group were analyzed for transgene expression 72-h postinfusion. In agreement with the in vitro data, we found that the Ad5 vector was not able to transduce MDA-MB435 cell-derived liver metastases in vivo, whereas it transduced hepatocytes efficiently (Fig. 4,B, Ad5 panel). As shown before, the Ad5/35 vector transduced only a small percentage of hepatocytes in mice (7 +/− 3%). Surprisingly, we found that the Ad5/35 vector efficiently transduced only occasional metastases (Fig. 4,B, Ad5/35 panel). From >3000 metastases counted on multiple sections, only 8 +/− 5% of metastases contained β-galactosidase-positive cells. These findings were confirmed with Ad5 and Ad5/35 vectors expressing GFP under the control of the MSCV promoter, which has been shown to be active in MDA-MB435 cells and mouse hepatocytes in vitro (Fig. 4 B, bottom panels). This demonstrated that the rare transduction of metastases was not linked to a specific promoter or transgene. To exclude that down-regulation of Ad35 knob-interacting receptor(s) in tumor cells accounted for the decreased susceptibility of tumor cells to Ad5/35 infection in vivo, tumor cells isolated from microdissected metastases (from tumor-bearing mice not infused with Ad) were infected with Ad5 or Ad5/35 vectors. Staining for β-galactosidase expression demonstrated that MDA-MB435 cells, recovered from liver metastases, were efficiently transduced by the Ad5/35 vector and refractory to Ad5 infection (data not shown).

A distinct morphological feature of MDA-MB435 cell-derived liver metastases was that tumor nests were encapsulated by an extensive ECM-containing collagen and laminin. It also appeared that the blood vessels rarely penetrated through the ECM capsule. Therefore, we hypothesized that the number of tumor-feeding blood vessels, the stage of their development, and/or the thickness of the ECM could be factors that limit the ability of Ad particles to infect cells otherwise susceptible to virus infection. To evaluate this, we more closely examined the structures of metastases that were susceptible to Ad5/35 infection. Histological analysis of serial sections stained with antibodies against β-galactosidase, laminin, and CD31 (Fig. 4C, top and bottom rows, left and middle panels) showed that >95% of metastases that expressed β-galactosidase were in close proximity to areas that stained positive for CD31. This suggests that contact of metastases with blood vessels is an important factor for efficient Ad transduction in vivo. Additionally, the majority of transduced metastases had visibly less laminin separating the blood vessels and tumor cells (Fig. 4,D, compare to Fig. 4 C, right panels). This indicated that the ECM represents a physical barrier for Ad infection of tumor cells.

The in vivo studies demonstrated that a capsid-modified Ad vector had reduced tropism to mouse liver cells after systemic application. However, the modified vector transduced only a small subpopulation of metastases with direct access to blood vessels. This suggests that efficient targeting of liver metastases not only depends on tumor specific tropism but is also affected by anatomical barriers limiting accessibility of tumor cells to Ad infection.

Retargeting of Ads is considered to be crucial for maximizing the number of initially transduced tumor cells and minimizing hepatocellular damage caused either by a direct toxic effect of expressed viral/oncolytic proteins or by CTL-mediated immune responses against transduced hepatocytes (33, 34).

In this study, we tested whether the new tropism of a capsid-chimeric Ad5/35 vector could be exploited to achieve tumor-specific targeting in vivo. Ad5/35 efficiently infected a CAR^low human breast cancer cell line that was refractory to infection with common Ad5-based vectors. This property of Ad5/35 is potentially important because it appears that many malignant tumors in patients have low levels of CAR expression (6). Because the level of CAR directly correlates with Ad5 infectivity (5, 35, 36), it is expected that the efficacy of routinely used Ad5-based vectors in these tumors will be limited. Administration (i.v.) of Ad5 vectors leads mainly to transgene expression in liver, which has been explained by a combination of vascularity, endothelial fenestrations, and high CAR levels on hepatocytes (29). In this context, another important feature of Ad5/35 is its low affinity to and infectivity of primary mouse hepatocytes.

The majority of solid malignant tumors found in patients are encapsulated by ECM (37). The presence of laminin and collagen around tumor clusters has been directly associated with their metastatic potential and subsequent poor clinical prognosis (15, 32). In our animal model, human breast cancer cells transplanted through the portal vein formed multifocal hepatic tumors surrounded by basement membranes, histologically mimicking an advanced tumor stage observed in humans.

Systemic application of Ad5/35 in our mouse tumor model resulted in transduction of only 8% of metastases. This selective transduction appeared to be because of differences in accessibility of the metastases to the virus. Our data indicated that vascularization of metastases increased transducibility, whereas ECM represented a physical barrier to Ad transduction. Important factors that account for the selective transduction of metastases may include varying discontinuities present in the basement membrane, which is considered to be more permeable in malignant tumors than in benign tumors (38), or heterogeneity in tumor-feeding blood vessels, including the density as well as the structure and fenestration of blood vessels (39). A potential role of the ECM in inhibiting the Ad infection in tumors has been suggested previously (29). It was also reported that local administration of proteases enhanced Ad-mediated gene transfer in arterial tissue (20) or gliomas (3).

Although only rare individual metastases were transduced by Ad5/35, the majority of tumor cells in these metastases demonstrated transgene expression. At the same time, the number of transduced hepatocytes was on average 10 times less than with the conventional Ad5 vectors. In this context, the Ad5/35 represents a vector retargeted to tumor cells. Considering this, the question may arise whether Ad5/35 is a relevant tumor specific vector for clinical trials. Our mouse model with human tumor cells engrafted in a mouse liver cannot answer this question. It also has to be noted that Ad5/35 can efficiently transduce human hematopoietic progenitor cells in vitro (27, 40). However, it is not clear whether i.v. application of Ad5/35 will result in transduction of bone marrow cells. Notably, Ad5/35 does not transduce peripheral blood cells in vitro⁴ or human fibroblasts in organoid cultures (41).

Our findings suggest that transduction of liver metastases not only requires tumor-specific Ad tropism but also new strategies to increase accessibility of tumor cells to systemically applied Ad vectors. Potential approaches to increase tumor transduction include stimulation/facilitation of trans-endothelial migration or digestion of the ECM, e.g., by the expression of laminin-specific matrix metalloproteinases or collagenases in stromal cells adjacent to tumors.

We thank Cheryl Carlson and Kathrin Bernt for helpful discussion.