Salmonella engineered to express CD20-targeting antibodies and a drug-converting enzyme can eradicate human lymphomas

Developing tumors are scrutinized by an immune system capable of recognizing and destroying aberrant cells before they become pathologic. To survive, a tumor must adopt a number of well-studied immune evasion mechanisms that overcome this monitoring.¹  Cancer immunotherapy seeks to harness the body’s immune response to treat tumors despite immune evasion adaptations. One interesting and innovative approach uses tumor-colonizing bacteria to counteract the immunosuppressive microenvironment. Indeed, a number of bacterial strains, including Clostridium, Bifidobacterium, and Salmonella, exhibit a remarkable tropism for tumors, indicating a therapeutic potential (reviewed by Forbes).²  Although the precise mechanisms facilitating bacterial tumor colonization are unclear, both the leakiness of the tumor vasculature and the hypoxic conditions found therein contribute.

Attenuated strains of Salmonella typhimurium represent valuable models for in vivo use, being genetically well characterized, easy to manipulate, and approved for administration in humans.³  Our group,^4,5  among others,^6,7  has previously demonstrated that Salmonella can eradicate tumors in mouse models. Indeed, intratumoral injection of Salmonella into aggressive melanomas has led to tumor regression and rejection. Furthermore, bacterial injection into primary tumors has caused significant growth retardation of untreated, distal tumors from Salmonella-dependent secondary immune responses created against tumor-associated antigens.⁴  This was reliant on the Salmonella-dependent upregulation of connexin43, a transmembrane protein involved in gap junction (GJ) formation.⁸  GJ between tumor cells and dendritic cells (DCs) favored tumor antigen crosspresentation and the establishment of an anti-tumor immune response.

In this study, we aimed to engineer Salmonella to reduce its systemic effects, while retaining its antitumor activity and potentiating its immune system–independent cytotoxic effects. Anti-tumor antibodies used alone or to convey cytotoxic cargoes (radiolabels, prodrug-converting enzymes) represent frontline therapies for many cancers.^9-11  We hypothesized that anti-tumor antibody expression would enhance Salmonella tumor cell invasion while disrupting general cell invasion. As a proof-of-concept experiment, we selected and characterized a camelid single-domain (VHH) antibody against the human CD20 (hCD20) tumor antigen from a phage-display library¹²  and expressed it on Salmonella. CD20, the antigen targeted in rituximab antibody immunotherapy of non-Hodgkin and diffuse large B-cell lymphomas,^10,13  represented an ideal target, as it was a well-studied tumor-associated antigen.

Here, we show success not only in specifically targeting tumor cells but also in delivering within them, via Salmonella infection, the herpes simplex virus thymidine kinase (HSV-TK). Thus, tumor cells became ganciclovir sensitive, thereby potentiating the therapeutic effect of Salmonella even in the absence of adaptive immunity.

B16-F10 murine melanoma cells were transfected with either hCD20-expressing or empty vector Babe-IRES-Puro (BIP) plasmid. Cells were harvested after 2 days, and hCD20 expression was verified by a fluorescence-activated cell sorter (FACS) using a rabbit anti-hCD20 antibody (Abcam). A total of 5 × 10⁶ B16F10-BIP empty vector cells were incubated with 1 × 10¹⁵ transforming units of the COGENTECH 1–naïve camelid VHH bacteriophage display library.¹²  Cells were pelleted, and supernatant (containing unbound phage) was added to 5 × 10⁶ B16F10-BIP-hCD20 cells. Phage elution, propagation, quantification and purification, and actual phage transforming units were performed as described previously,¹²  and output phage was used for second and third pannings.

A total of 200 third-round B16F10-hCD20–interacting phage colonies were picked, and VHH-pIII-protein–containing periplasms were produced as described previously.¹²  Clones were screened in quadruplicate by dot blot vs either CD20⁺ Namalwa B lymphoma or CD20⁻ 697 B leukemia cells. Binding was detected via anti-HA 12CA5 Ab [1 μg/mL] and goat anti-mouse-HRP Ab, followed by enhanced chemiluminescence development (GE Healthcare Life Sciences), and was quantified using ImageJ software. A total of 17 clones most consistently binding CD20⁺, and not CD20⁻, cells were sequenced, revealing repeated VHHs (5/17), of which clone 2G9 (3/17) alone could be purified.

2G9- and glutathione S-transferase (GST)b1-VHH were cloned into pHEN6 and VHH purified as described previously.¹²  FACS verifications of VHH-binding specificities were detected via anti-His mAb (Santa Cruz) followed by sheep anti-mouse-biotin IgG-PE antibodies (Sigma-Aldrich) and were compared with cells probed with rabbit anti-CD20 antibody (Abcam), followed by rat anti-rabbit AlexaFluor546 (Molecular Probes).

2G9-CD20 protein interaction was verified by coincubating VHH with Namalwa cell membrane fractions (Native Membrane Protein Isolation kit [Calbiochem]). Antibodies were pulled down (nickel nitrilotriacetic acid agarose [VHH] or Protein A-Sepharose [rAb]), washed, and boiled into Laemmli buffer for western blot analysis. The presence of CD20 in the immunoprecipitates was visualized as described above.

All experiments used the SL3261 aroA–deficient S typhimurium (AT-Salmonella)¹⁴  as described previously.^4,5  Colony-forming units (cfus) were determined for each transformant with or without isopropyl-L-thio-beta-D-galactopyranoside induction and were verified after each experiment.

hCD20 was amplified by polymerase chain reaction from Ramos human B lymphoma cDNA, sequence verified, and cloned into the BIP (retroviral)¹⁵  and 945-green fluorescent protein (GFP) (lentiviral) vectors (Dr L. Naldini, Telethon Institute of Gene Therapy, San Raffaele Scientific Institute, Milan, Italy).¹⁶  The HSV-TK gene from pHSV-TK (Dr V. Russo, Department of Biological and Technological Research, San Raffaele Scientific Institute)¹⁷  was cloned into pET15b plasmid. VHHs were cloned into the pHEN6¹²  and pVUB4¹⁸  vectors. Primer sequences and detailed cloning strategies are available on request.

Outer membrane translocation of VHH-ompA fusion proteins were routinely verified by dot blot of intact bacteria and were detected by western blotting. Bacterial outer membrane protein fractionation was performed as described previously,¹⁹  and VHH was similarly detected.

B16F10 cells were cultured as described previously,⁴  as were the MCA203 fibrosarcoma²⁰  and the CT26 colon carcinoma²¹  cells (Dr Mario Colombo, Immunotherapy and Gene Therapy Unit, Istituto Nazionale dei Tumori, Milan, Italy).

hCD20-945-GFP and 945-GFP empty vector lentivirus-containing supernatants were produced, and infections were performed following standard protocols. FACS verification of hCD20-expression was performed as described above.

Cells were counted, seeded onto Permanox chamber slides (Laboratory-Tek), and coincubated with Salmonella at defined multiplicities of infection (MOIs) for 2 hours at 37°C. Slides were then subjected to immunofluorescence staining to discriminate extracellular (doubly positive for Cy5- and Cy3-secondary antibodies) and intracellular bacteria (Cy5-singly-positive)²²  using an anti-Salmonella antibody (Virostat). Cytoplasm was visualized via phalloidin-fluorescein isothiocyanate (FITC) and nuclei by 4′,6 diamidino-2-phenylindole. Images were taken on an Olympus AX70 fluorescent microscope. Quantification of extracellular and intracellular bacteria was performed in a blinded manner.

Six-week-old female C57/Bl6J-OlaHsd mice (Harlan Laboratories) were injected subcutaneously with 5 × 10⁵ MCA203-empty vector or -CD20-GFP cells in contralateral flanks. Tumors were grown until the desired dimensions were reached: short-term analyses 5 to 8 mm²; long-term analyses 3 to 4 mm².

For biodistribution experiments, tumor-bearing mice received a single intravenous injection of 5 × 10⁶ cfus of the indicated Salmonella. Animals were euthanized 24 hours later; liver, spleen, tumor, bone marrow, and brain were excised, minced, and digested with collagenase D (Roche). Total bacterial load per organ was quantified by serially diluting counted cells. Intracellular bacteria were determined in parallel by a gentamicin-protection assay (50 μg/mL, 1 hour, 37°C).

For short-term treatments, mice were either immunized or were not immunized vs Salmonella via oral gavage¹⁹  and then received intravenous 5 × 10⁶ cfus of the indicated Salmonella. Tumor dimensions were measured every 2 days for up to 2 weeks.

For long-term survival analyses, 1 × 10⁸ cfus were administered intratumorally into nonimmunized mice. Tumor dimensions were measured as above, and mice were monitored for up to 6 months. Mice were judged tumor-free when no mass was palpable for 1 month minimum. Surviving mice were rechallenged with 5 × 10⁵ MCA203 cells at a distal site, and tumor development was monitored for 3 months.

Five-week-old, female nonobese diabetic/severe combined immunodeficiency (NOD/SCID) mice (Charles River) were injected subcutaneously with either 5 × 10⁶ Namalwa or Karpas299 cells.²³  Bacterial biodistribution was determined as detailed above. When tumors had grown sufficiently, mice received 1 × 10⁸ cfus of bacteria intratumorally once per week for 2 weeks, and tumor growth was tracked as mentioned.

For ganciclovir sensitization experiments, Salmonella containing both the pVUB4KAN (with or without VHH) and pET15b-HSV-TK plasmids was selected. HSV-TK expression was verified using goat anti-HSV-TK antibody (SantaCruz). Bacteria were injected either intratumorally (1 × 10⁸ cfus) or intravenously (5 × 10⁶ cfus) into tumor-bearing animals every 4 days until the end of the experiment; control animals received phosphate-buffered saline (PBS) alone. Concomitantly, mice were injected intraperitoneally with 50 mg/kg of ganciclovir every 12 hours for the duration of the experiment (100 mg/kg/day).²⁴  Tumor growth was monitored every 2 days. Survivors partially or fully rejecting tumors were treated for 1 additional month.

All animal studies were approved by the Institutional and Italian Health Ministries Animal Care and Use Committees. All mice were maintained and handled in accordance with local and international laws and relevant regulations.

The tropism of Salmonella for tumors has been well documented.^3,25,26  Although usage of attenuated strains, such as the AT¹⁴  mutant, significantly reduces acute effects,²⁷  its widespread, general cell invasion dilutes its therapeutic efficacy.^26,28  We hypothesized that Salmonella could be engineered to become tumor specific via the bacterial surface overexpression of an antibody against a tumor-associated antigen. Thus, antigen-expressing cells (tumors) would be invaded, but the antigen-negative cells (healthy tissue) would not. As a proof-of-principle experiment, we chose the hCD20 protein, the target of the rituximab anti-lymphoma antibody immunotherapy.^10,13  To specifically fish out CD20-directed VHH and not general lymphoma-associated targets, we transformed B16F10 melanoma cells to ectopically express hCD20. Specific VHHs recognizing hCD20 in its native conformation were isolated by panning a preimmune phage-display library¹²  against intact cells. Despite poor protein expression by the B16F10 cells (7% hCD20-positivity [supplemental Figure 1A]), we observed a significant enrichment of recovered phages after second and third rounds of panning (13-fold and 1.9-fold, respectively; data not shown), compared with negative control pannings. A total of 200 clones were then selected and were screened, as described in the “Methods” section. The hCD20-specificity of a selected number of clones (n = 17) was confirmed via FACS, examples of which can be seen in Figure 1A and supplemental Figure 1B. Sequencing revealed 2 VHH-sequences present in multiple clones, and they were subcloned and purified for further characterization. Because of its biochemical stability, we selected the VHH encoded by clones 2G9/1H5/2B10 (2G9), which was able to distinguish by FACS, compared with a GST-binding control VHH (GSTb1), CD20⁺ lymphomas from CD20^- leukemias (Figure 1B), or CD20⁻ T lymphomas (Karpas299; data not shown), and effectively immunoprecipitated hCD20 protein from cell lysates (Figure 1C).

To create Salmonella displaying 2G9 on its outer membranes, the VHH was cloned into the pVUB4 vector, fusing it to the ompA transmembrane protein domain.¹⁸  The isopropyl-L-thio-beta-D-galactopyranoside-induced expression of 2G9 on the outer surface of AT-Salmonella was visualized using purified membrane fractions (Figure 1D), and VHH translocation was thereafter monitored by dot blot (data not shown). As a final confirmation, 2G9-Salmonella efficiently bound hCD20⁺ Namalwa cells, although failing to bind CD20⁻ 697 cells, as visualized by immunofluorescence (supplemental Figure 2).

Next, we determined whether VHH expression could direct bacterial invasion to cells expressing the targeted marker on their membranes, as AT-Salmonella remains competent to invade murine cells. We generated mouse colon carcinoma (CT-26) cells (Figure 2A), and fibrosarcoma (MCA203) cells (images not shown) to stably express hCD20. CD20⁺ and CD20^- WT counterpart cells were coincubated with AT-Salmonella expressing no VHH, 2G9-VHH, or the GSTb1-VHH at defined MOIs. Salmonella localization was then examined via immunofluorescence under conditions allowing for the discrimination of intracellular and extracellular bacteria, based on differentially staining before and after permeabilization.²²  As shown in Figure 2A-B, 2G9-Salmonella bound and invaded the CD20+ cells at least as efficiently as unmodified AT-Salmonella, but significantly more than GSTb1-Salmonella (displaying the irrelevant control VHH). 2G9-Salmonella bound cells 5.6-fold (MCA203) (P = .006) and 7.9-fold better (CT26) (P = 3 × 10⁻⁹) than the GSTb1 control (Figure 2B). As hypothesized, that the surface-expression of VHH on Salmonella may restrict its infectivity, expression of GSTb1-VHH inhibited the capacity of Salmonella to bind and invade tumor cells, independent of the presence of CD20. Indeed, bound 2G9-Salmonella were significantly more invasive than GSTb1-Salmonella (64% vs 29% for MCA203 cells [P = .0007] and 49% vs 16% for CT26 cells [P = 7 × 10⁻⁶]) (Figure 2C). The cfus for each strain were verified postinfection (Figure 2D), thus eliminating the possibility that the observations were the result of differing amounts of bacteria present. Altogether, these results suggest that bacterial surface expression of recombinant antibodies could disrupt nonspecific cell interactions of Salmonella while the antibody-mediated recognition of a cellular marker preserves cellular binding and invasion.

We then confirmed that CD20-targeted Salmonella lost the capacity to bind or invade cells lacking CD20, via infections performed with CD20⁻ MCA203 (Figure 3A) and CT-26 (Figure 3B) cells. Although wild-type AT-Salmonella entered both CD20⁺ and CD20⁻ cells with similar efficiency (compare Figure 2A with Figure 3A-B, left panels), 2G9-Salmonella predominantly invaded the CD20⁺ cells (Figure 2A-B and Figure 3A-B). Indeed, when encountering cells lacking their targeted antigen, 2G9-Salmonella mimics GSTb1-Salmonella, becoming incapable of binding or invading cells (Figure 2A-B and Figure 3A-B). Consequently, the AT-Salmonella bound the MCA203 and CT-26 cells 4.4-fold (P = .01) and 1.8-fold (P = .01) better, respectively, than its 2G9-expressing counterpart (Figure 3C). AT-Salmonella also invaded better (67% vs 33% [MCA203, P = .02] and 59% vs 21% [CT26, P = .002], AT vs AT-2G9, respectively) (Figure 3D) when using comparable cfus in each infection (data not shown). These data confirm our hypothesis that overexpression of an antibody at the Salmonella surface confines binding and invasion only to cells expressing the targeted antigen.

Next, we examined the capacity of the 2G9-Salmonella to invade tumors in vivo. Mice were injected subcutaneously in opposite flanks with MCA203 cells infected with either empty vector or hCD20-encoding vector, with the empty vector tumors acting as hCD20-free negative control tissues to confirm antigen-specific targeting or invasion in vivo. When the tumors had grown sufficiently, the mice were injected intravenously with 5 × 10⁶ cfus of either the 2G9-Salmonella or the GSTb1-Salmonella. At 24 hours later, the animals were euthanized and bacterial distribution and invasiveness were quantified via a gentamicin protection assay. Although the total bacterial load within either the CD20⁺ or CD20⁻ tumor was equivalent, regardless of which VHH was displayed (data not shown), the 2G9-Salmonella accumulated intracellularly 5.2-fold more within the CD20⁺ tumor than the CD20⁻ tumor within the same animals (Figure 4A) (P = .04), indicating that VHH-targeted Salmonella did not lose tumor tropism and retained antigen-specific invasion in vivo. Importantly, only minimal in vivo invasion of either tumor was detected using GSTb1-Salmonella (Figure 4A) (P < .05). Importantly, compared with mice receiving unmodified AT-Salmonella, VHH-targeted bacteria did not exhibit significant off-target invasion of known in vivo reservoirs (spleen and liver)²⁶  (Figure 4B), did not aberrantly localize elsewhere (data not shown), and did not affect the number of circulating B cells (supplemental Figure 3), thereby confirming the disruption of general, nonspecified cell invasion in vivo.

To determine whether antibody-dependent tumor recognition and invasion by Salmonella could be therapeutically beneficial, we orally immunized mice against Salmonella,¹⁹  and CD20⁺ and CD20⁻ MCA203 tumors were intratumorally injected with 1 × 10⁸ cfus of either 2G9- Salmonella or empty vector-Salmonella. 2G9-Salmonella elicited a 50% reduction of CD20⁺ tumors with respect to their preinjected state, but the CD20⁻ tumors continued to grow (Figure 4C-D, P = .01 at day 10). In contrast, treatment with the empty vector-Salmonella failed to induce tumor regression, which, instead, doubled in size within 10 days of initial treatment of the tumors (Figure 4D), much like the 2G9-injected CD20⁻ tumors. As expected,⁴  intratumoral Salmonella was itself partially protective, as PBS-injected tumors grew faster than Salmonella-injected tumors (Figure 4D). Empty vector-Salmonella and not GSTb1-Salmonella served as a control because the latter, being incapable of infecting tumors, would have masked the nonspecific anticancer effect of invasive Salmonella, thus confounding the comparison.

Although intratumoral delivery of Salmonella into preimmunized animals was protective in an acute response, whether the 2G9-Salmonella–induced regression of CD20⁺ tumors could increase survival duration remained uncertain. Nonimmunized mice bearing either CD20⁺ or CD20⁻ MCA203 tumors were injected intratumorally, twice, 1 week apart, with either the 2G9-Salmonella or empty vector-Salmonella (1 × 10⁸ cfus), and tumor dimensions and mouse survival duration were then tracked. As seen in Figure 4E, overall survival time was significantly extended in only those mice bearing hCD20 tumors and receiving 2G9-Salmonella (log rank P = .02). Indeed, we consistently observed that up to 67% of the CD20⁺ tumor-bearing mice treated with 2G9-Salmonella survived tumor-free for more than 3 months after the end of the experiment. This rate was significantly better than the rate seen for CD20⁺ tumors treated with empty vector-Salmonella (12.5%), whereas no mice bearing CD20⁻ tumors exhibited prolonged survival time with either bacterium. Successful tumor rejection generated anti-tumor immunity, as subsequent rechallenge of the CD20⁺ tumor survivors (n = 6) with MCA203 cells failed to yield secondary tumors (data not shown). The number of cfus used within each condition was verified to be identical (Figure 4F and data not shown). These observations demonstrate that the expression of tumor-antigen–specific VHH on Salmonella improves the selectivity for antigen-bearing tumor cells, culminating in tumor regression, rejection, and the establishment of tumor-protective immunity while simultaneously disrupting nonspecific invasion of healthy tissues.

We further sought to extend the applicability of our system by examining whether our tumor-targeting Salmonella could be used to treat human tumors. Therefore, we used the NOD/SCID xenograft model in which the absence of an adaptive immune system allows for human tumor cell outgrowth. CD20⁺ Namalwa cells or the CD20⁻ Karpas299 were injected subcutaneously in opposite flanks²³  and then were injected 3 weeks later intratumorally with 1 × 10⁸ cfus of either 2G9-Salmonella or empty vector-Salmonella. Notably, the 2G9-bacteria invaded the CD20+ tumor cells by an average of 16.9-fold better than the CD20⁻ tumors (P = .03), and 5.5-fold better than the empty vector-Salmonella (P = .05) (Figure 5A). Despite this finding, no significant tumor regression was observed after administration of either the 2G9-Salmonella or empty vector-Salmonella according to the previously outlined immunotherapeutic protocol (Figure 5B). This trend was to be expected, as we had observed previously that the formation of tumor-specific CD8+ T cells, lacking in NOD/SCID animals, was essential for Salmonella-based cancer immunotherapy.⁴ 

The absence of adaptive immunity allowed us to therefore evaluate the usefulness of Salmonella to intracellularly target tumor cells for the delivery of a prodrug-converting enzyme. Thus, we generated 2G9-Salmonella expressing the HSV-TK enzyme (data not shown). This would allow us to deliver HSV-TK only into CD20-expressing cells, by means of the 2G9-Salmonella infection, thus sensitizing them to the prochemotherapeutic, ganciclovir, which is converted into a deoxyguanosine triphosphate analog that induces apoptosis on DNA incorporation.¹⁷  NOD/SCID animals bearing Namalwa tumors received 1 × 10⁸ cfus of either 2G9-AT-HSV-TK, empty vector-AT-HSV-TK, 2G9-AT, or PBS intratumorally every 4 days with or without intraperitoneal ganciclovir injection [100 mg/kg/day].²⁴  Only the 2G9-AT-HSV-TK, in combination with ganciclovir, delayed tumor growth or initiated regression (Figure 5C). Ganciclovir alone or combined with empty vector-HSV-TK-Salmonella failed to slow tumor growth. As seen in Figure 5C, average tumor dimensions were significantly smaller for the 2G9-AT-HSV-TK-ganciclovir combination therapy, compared with all other conditions tested (day 12 compared with empty vector-HSV-TK-ganciclovir [P = .02] or PBS-ganciclovir [P = .0001]). Likewise, the clinical response for this combination was vastly better with 2 of 5 animals exhibiting delayed tumor growth and 2 of 5 showing tumor reduction, 1 of which completely rejected the tumor. Remarkably, in more than 50 NOD/SCID tumor-bearing animals treated with PBS-ganciclovir alone, empty vector-AT-HSV-TK-ganciclovir, or 2G9-HSV-TK or empty vector-AT-HSV-TK alone, not a single animal exhibited delayed tumor growth or tumor regression. Actual cfus were verified to be identical among all tested conditions (Figure 5D). Although this result was quite promising, its usefulness is limited to only those tumors easily discovered and injected. Therefore, we sought to deliver the engineered Salmonella intravenously and track its effect on tumor growth. Strikingly, 2G9-AT-HSV-TK was quite successful in mediating significant reduction in tumor volume when coadministered with ganciclovir in 86% of the treated mice (Figure 5E). All other treatment regimens (2G9-AT-ganciclovir, empty vector-AT-HSV-TK-ganciclovir, or PBS-ganciclovir) failed to delay tumor progression. Similar results were obtained by intravenous administration into immune-competent mice bearing CD20+MCA203 tumors (Figure 5F). Again, 2G9-AT-HSV-TK-ganciclovir treatment induced tumor shrinkage in 71% and, remarkably, complete tumor ablation in 57% of the treated animals. Additionally, 2G9-AT-ganciclovir in this case achieved tumor reduction in 50% of treated animals before relapse occurred. Again, no animals receiving empty vector-AT-HSV-TK-ganciclovir or PBS-ganciclovir had significant delays in tumor growth. Therefore, the data indicate that we have created a novel therapeutic strategy allowing for the targeting and rejection of human tumors regardless of the presence of a functional adaptive immune response by using tumor-targeted Salmonella carrying a prodrug-converting enzyme with a systemically administered prochemotherapeutic agent.

Survival of neoplastic cells despite constant immune monitoring is a key step that is acquired during tumorigenesis. The mechanisms comprising tumor immune evasion pose significant challenges to the application of immunotherapies. Indeed, it has proven difficult to boost preexisting anti-tumor immunity or generate it de novo in patients exhibiting widespread disturbances in the immune system. Therefore, new therapeutic approaches that circumvent immune escape are critical. The efficacy of Salmonella to target and eradicate growing tumors has been well documented.^2,4,5  However, colonization and persistence of Salmonella within healthy tissues weaken its therapeutic efficacy and causes undesirable adverse effects.^3,26  Therefore, it became necessary to develop novel methods to boost the tumor tropism of Salmonella.

This development was accomplished in our current study via the surface expression of a single-domain antibody vs a tumor-associated antigen (hCD20). Bacteria overexpressing cell surface 2G9 efficiently bound and entered CD20⁺ MCA203, CT26, and Namalwa cells, whereas Salmonella overexpressing GST-specific VHH failed to appreciably do so. Uptake of 2G9-Salmonella was dependent on the recognition of membrane-expressed hCD20, culminating in the remission and rejection of CD20⁺ tumors. Whereas others have previously used concomitant delivery of bispecific antibody and Salmonella mutants,²⁹  or tumor-targeting VHHs³⁰  to facilitate tumor rejection, our novel methodology uses viable bacteria engineered to express all moieties required for tumor destruction. Thus, we can create patient-specific “intelligent missiles” delivering a variety of cytotoxic payloads into targeted tumor cells, thus facilitating their specific destruction.

Accordingly, we found that the expression of VHH molecules almost completely ablated the generic cell infectivity of Salmonella in vitro and in vivo. Diminished nonspecific cell invasion strongly reduced the undesirable accumulation of Salmonella within mouse spleens and livers, thereby boosting their therapeutic usefulness. As has been proposed elsewhere,¹⁸  a fuller understanding of how bacteria colonize healthy and tumor tissues could give insight into how the tumors themselves grow and metastasize. Additionally, it would be interesting to assess whether enhanced tumor targeting and invasion may disfavor biofilm formation that could otherwise diminish anticancer activity of Salmonella.³¹ 

These findings represent a clear improvement toward the control of in vivo–delivered bacteria. The highly attenuated salmonellae currently approved for in vivo use, by necessity, lack bacterial-dependent cellular cytotoxicity.⁴  Rather, these vectors depend on the patient’s immune system to recognize and destroy infected cells. Indeed, we have recently shown that systemically administered Salmonella mediated an anti-tumor response via the DC upregulation of connexin43, thereby mediating GJ-dependent transfer of tumor antigens into recruited DCs, which, in turn, educate tumor-specific CD8+ T cells to reject tumors.⁸  Such a strategy may work for patients with a robust immune system, a condition often missing in advanced cancers. The ability to load tumor-targeted Salmonella with cytotoxic cargos may augment the immunostimulatory activity of Salmonella with a cytotoxic response specifically within tumor cells. Therefore, we engineered 2G9-Salmonella expressing the HSV-TK enzyme, which imparted ganciclovir sensitivity to infected cells. Accordingly, only CD20+ human xenografts cotreated with 2G9-Salmonella and ganciclovir underwent tumor reduction or rejection (Figure 5D-E). A similar system has previously been used to infect and visualize tumors via the HSV-TK–dependent incorporation of radionucleotides.³²  The successful disruption of human tumor growth in NOD/SCID animals indicates that cytotoxicity occurs independently of adaptive immunity and may thus be a valuable tool to treat immunocompromised patients without safety concerns. Furthermore, our observation that antibody-targeted Salmonella can specifically enter lymphoma cells while being rapidly cleared from healthy tissues strongly suggests its usefulness for treatment of blood-borne cancers, a property hitherto unattributed to bacterial immunotherapies.

In conclusion, we have developed a novel methodology to generate cell/tumor-specific invasive Salmonella via the expression of VHH antibodies on their outer surface. Consequently, Salmonella entry occurs only into cells displaying the target antigen, abolishing generic bacterial cell entry both in vitro and in vivo. The effectiveness of targeted bacteria coexpressing HSV-TK, delivered either intravenously or intratumorally, indicates that the bacteria can be engineered to transport cytotoxic molecules into tumors and metastases, destroying them in the absence of adaptive immunity. Similarly, Salmonella can be loaded with radioisotopes,³²  short hairpin RNAs,^7,8  or tumor-exclusive immunogenic antigens^33,34  to maximize tumor-specific killing.

Our work, therefore, represents a significant advancement toward the creation of safe, therapeutically efficacious Salmonella for the treatment of both detected and undetected tumors in humans. Furthermore, the ease with which specific VHHs can be isolated can streamline the creation of patient-specific Salmonella vectors via high-throughput screening for VHHs preferentially engaging patient tumors.

The authors thank Drs Luigi Naldini, Vincenzo Russo, and Kenneth B. Marcu for sharing their plasmids and Dr Mario Colombo for the gift of the MCA203 and CT-26 cell lines.

M.R. is supported by grants from the European Commission (ERC, Dendroworld), the Italian Association for Cancer Research (AIRC), the Association for International Cancer Research (AICR) and the Ministry of Health (Ricerca Finalizzata).

Contribution: P.E.M. performed and designed research, analyzed data and statistics, and performed manuscript preparation; A.P. collected data and analyzed data and statistics; A.M. performed experimental design and contributed vital new tools; A.d.M. performed research design and manuscript editing; and M.R. performed research design, data analysis, and manuscript editing.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Maria Rescigno, Department of Experimental Oncology, European Institute of Oncology, Via Adamello 16, 20139 Milan, Italy; e-mail: maria.rescigno@ieo.eu.