High-Throughput Screening for Salmonella Avirulent Mutants That Retain Targeting of Solid Tumors

Salmonella enterica serovar Typhimurium is a facultative anaerobic bacterium that infects a wide variety of animal hosts and naturally accumulates in most solid murine tumors versus normal murine tissues at a ratio of 1,000:1 (1). Avirulent Salmonella mutant strains have been used to directly kill tumors (2–4) or to deliver constitutively expressed therapeutic proteins to tumors in mouse models (5–7). The use of Salmonella promoters preferentially active in tumors (8) should further increase the specificity and hence the safety of therapeutic systems. Avirulent Salmonella mutants used in cancer therapy have typically impaired biosynthesis of aromatic compounds, purines, or amino acids and were generally selected for their avirulence. For example, an aroA aroD double mutant was used to deliver the Flt3 ligand to treat melanoma in mice and resulted in 50% tumor regression (7). It is possible that a different avirulent mutant that grows better in tumors might have resulted in more complete tumor regression. The mutants A1 and A1-R deficient in leucine and arginine biosynthesis was effective against prostate cancer, breast cancer, pancreatic cancer, and osteosarcoma in nude mice (4, 9–11), yet A1-R mutations that contribute to tumor killing are unknown. Another Salmonella strain, VNP20009, has mutations in the purI and msbB genes resulting in modified purine biosynthesis and reduced septic shock potential. VNP20009 is widely used for delivery of cancer therapeutics in mice (12). However, it showed only moderate tumor targeting when tested in clinical trials of human patients with metastatic melanoma (13). Whereas some of these mutants have shown a therapeutic value in cancer, the true potential of Salmonella may fully be explored if all Salmonella mutants were tested individually. However, this would be practically impossible and prohibitively expensive. Here, we describe a high-throughput fitness screening of Salmonella mutants in all nonessential genes to determine which mutants were the best at accumulating in tumors while being disabled for growth in normal tissues. Mutants with reduced fitness in normal tissues but with unchanged fitness in tumors were identified and have potential use as cancer therapeutics.

Specific knockout mutants, described in Table 2, were generated in the S. Typhimurium strain 14028 background using the Lambda-Red recombination method with modifications (14). A transposon library in 14028 was constructed using the EZ-Tn5 <T7/KAN-2> kit (Epicenter). The library had ∼41,000 kanamycin-resistant mutants.

PC-3 human prostate cancer cells and MDA-MB-435 human breast cancer cells⁷

A frozen aliquot of the initial library was used to inoculate 100 mL of LB. After overnight growth, bacteria (input library) were pelleted and washed three times with PBS, and 10⁷ colony-forming units (cfu) were injected intratumorally (i.t.) into twelve 6-week-old nude mice (six mice bearing subcutaneous human PC-3 prostate, six mice bearing subcutaneous human MDA-MB-435 breast cancer) and i.v. into three tumor-free nude mice. Two days after injection, tumors and normal tissues (spleen, liver, and lung) from tumor-free mice were recovered and homogenized in PBS. An aliquot was plated on kanamycin-containing LB plates to determine the cfu. The remainder of the sample was added to kanamycin-containing LB and incubated overnight at 37°C (output libraries). The DNA adjacent to transposon insertions in library samples was amplified as described (16), with the following modifications: PCR amplifications were carried out using primers DOPR2 (CAACGCAGACCGTTCCGTGGCA) and CCT₂₄VN (CCTTTTTTTTTTTTTTTTTTTTTTTTVN). Nested PCR amplifications were carried out using primers CCT₂₄VN and KAN2FP1-B (GTCCACCTACAACAAAGCTCTCATCAACC). Further details of the experimental method and data analysis are presented in Supplementary Data 2A and B.

A library of 41,000 Salmonella mutants containing mini-Tn5 transposon insertions was constructed and pooled. The pool was injected into six human prostate (PC-3) and six breast-cancer (MDA-MB-435) tumors growing subcutaneously in nude mice and injected i.v. into three tumor-free mice. Bacteria were recovered after 2 days from tumors and from spleens, livers, and lungs of tumor-free mice.

During in vivo selection, defective mutants in genes contributing to fitness in the selective environment are lost from the pool. Differences in the mutant pool composition before (input pool) and after selection (output pool) can be detected using breast cancer microarray hybridization: Transposons were used that carry the T7 promoter sequence, allowing the specific amplification of genomic sequences adjacent to each insertion, which are then mapped on the Salmonella genome using a gene microarray (Supplementary Data 2A and B). The present study revealed two distinct classes of mutant phenotypes (Table 1; Supplementary Data 1).

This class contains mutants with reduced fitness in normal tissues (spleen, liver, and lung) and unchanged fitness in tumors. We identified mutants affecting at least 19 distinct genes within the SPI-2 island (i.e., ssrA, ssaB, ssaC, ssaD, sseB, sscA, sseC, sseE, ssaJ, STM1410, ssaK, ssaL, ssaM, ssaV, ssaN, ssaP, ssaQ, yscR, and ssaT). In addition, mutants in genes involved in a number of cellular functions were identified (Table 1). These include htrA, phoP, sifA, and a hypothetical operon composed of a putative acetyl-CoA hydrolase (STM3118), a putative monoamine oxidase (STM3119), and two putative transcriptional regulators (STM3120 and STM3121). Many of these mutants have previously been observed to be associated with fitness in spleen (13, 14). The observation of a similar effect on fitness in liver and lung is new but not unexpected. The fact that these mutants remain fit in tumors relative to other mutants is of potential practical importance for Salmonella use as a therapeutic delivery vector.

This class contains mutants with reduced fitness both in normal and in tumor tissues. Three mutants of the same operon involved in the synthesis of aromatic compounds were identified: aroM, aroD, and aroA. Previous reports describe the use of aroA and aroD mutants in cancer therapy (7, 17, 18). Mutants in genes related to lipopolysaccharide biosynthesis were also identified in this class (e.g., rfbK, rfbM, rfbC, and rfaQ). Whereas class 2 mutants are either avirulent or of reduced virulence, their impaired growth in tumors relative to class 1 mutants makes them less suitable for cancer therapy. The ability of mutants to directly kill tumors was not tested because our screen was designed only to identify mutants with reduced fitness in spleen but unchanged or improved fitness in tumors. Regardless of any ability to kill tumors, such mutants will be able to deliver and express therapeutics under the control of tumor-specific promoters.

We constructed individual gene knockouts of class 1 and class 2 Salmonella genes using the Lambda-Red recombination technique (19). These deletion mutants were tested for virulence in immunocompetent mice. Each mutant was injected (10⁵ cfu) i.v. into five C57BL/6 mice, which are particularly sensitive to Salmonella infection, and are thus a stringent test of virulence. Additional deletions were made in three genes that had no observable fitness phenotype in our microarray screen and used as controls (STM1459, ybjN, and feoB). Each mutant strain was assigned to one of three categories based on virulence level: virulent, mildly attenuated, or severely attenuated (Table 2). Virulent strains cause distress, dehydration, and death within 2 days after inoculation, which was observed with all three control mutants. In contrast, class 1 and class 2 mutants were either mildly or severely attenuated. Exposure to the mildly attenuated mutants STM3119, rfbI, rfaQ, rfbK, rfbM, sifA, and phoP caused signs of distress 2 to 6 days after inoculation (Table 2). The C57BL/6 mouse strain also presents similar symptoms within 2 days following the same dose of wild-type Salmonella administration (data not shown). All other mutants were severely attenuated and did not cause any distress during the 2-week experiment. These strains include SPI-2, htrA, STM3120, aroA, and aroD (Table 2). Assays with SPI-2 and STM3120 mutants were repeated three times, each with five mice. No signs of distress or death were observed for 4 weeks, after which all mice were sacrificed.

In summary, fitness assays on pools of mutants can be used as a primary screen for candidate attenuated mutants while simultaneously monitoring the relative ability to survive in tumors. The ability to screen thousands of candidates and evaluate individual mutants in parallel using arrays, and in the future, high-throughput sequencing, offers a clear advantage over conventional screening methods.

The competitive fitness of the class 1 mutant STM3120 in the tumor environment was tested in vivo against the class 2 mutants aroD and aroA. Input bacterial mixes were prepared in an approximately 1:1 ratio of STM3120 to aroD or aroA. Approximately 10⁶ cfu of the mixture were injected into human PC-3 tumors growing subcutaneously in nude mice. Input ratios were compared with output ratios recovered from tumor biopsies 2, 4, and 7 days after injection.

Two days after injection, the level of viable aroD mutants was 2 logs lower than that of STM3120. After 1 week, aroD was undetectable whereas STM3120 counts increased slightly. When STM3120 was competed against aroA, the aroA count was initially reduced by about 2 logs 2 days after injection but maintained at the same level 1 week after injection (1–2 logs less than STM3120). These results suggest that STM3120 outcompeted aroD to a greater extent than it outcompeted aroA (Fig. 1), consistent with the microarray data showing that aroA is more fit in tumors than aroD, but not as fit as STM3120 (Table 1).

The tumor-targeting capability of STM3120 was tested in five 6-week-old BALB/c mice bearing 4T1 breast tumors grown orthotopically for 10 days. Mice were gavaged with 7 × 10⁸ cfu of STM3120. Tumor biopsies were taken 2, 5, 7, and 9 days later⁸

These results suggest that intragastric delivery of STM3120 allows a sufficient number of bacteria to target and multiply in the tumor environment to levels (about 10⁷–5 × 10⁸ cfu per tumor) previously shown to effectively reduce tumor size after i.t. or i.v. injection (14). This is of importance because intragastric delivery of therapeutic strains may offer improved safety over i.v. delivery. A similar finding was recently made by Jia and coworkers (20), showing a significant anticancer effect of orally administered VNP20009 to mice bearing syngeneic subcutaneous B16F10 melanoma and Lewis lung carcinoma. Tumor targeting after oral delivery to BALB/c mice is likely due to the ability of S. typhimurium to produce a typhoid-like systemic infection in this mice strain. The ability to efficiently enter the bloodstream of humans after oral delivery may require additional engineering or the use of strains that are naturally efficient at this transition, such as Typhi strain Ty21.

The original screenings lasted only 2 days, the time necessary to ensure sufficient complexity in the mutant library without random loss of mutants that would confound the high-throughput analysis. As a result, mutants such as STM3120 were selected based on the fitness phenotype and not the tumor killing targeting. Nevertheless, we occasionally observed necrosis in tumors, likely associated with the injection of STM3120.

We have shown that high-throughput screening of a pool of transposon mutants allows the identification of novel Salmonella mutants with potential therapeutic value and the reevaluation of those previously used in cancer therapy. Mutants that retain tumor-targeting while being poor colonizers of normal tissue are candidates for delivery of cancer therapeutics. However, mutants will need to be tested in the intended host before the best candidates can be determined. Such approaches can be adapted to any host and tumor model and a wide variety of bacterial species.

The authors are pursuing the commercial application of engineered infectious agents to cancer therapeutics. M. Zhao and R.M. Hoffman are affiliated with AntiCancer, Inc. M. McClelland and N. Arrach are the founders of Vivocure, Inc.

We thank Steffen Porwollik, Brian Ahmer, and Helene Andrews-Polymenis for helpful discussions, Rocio Canals for help with library construction, and Charlene Cooper for her support and administrative assistance.

Grant Support: NIH grants R01AI034829, R01AI052237, and R21AI057733 (M. McClelland), Tobacco-Related Disease Research Program grant 16KT-0045 (N. Arrach), DOD grant W81XWH-08-1-0720 (M. McClelland), PBCT-CONICYT (Chile) and The World Bank grant ADI-08/2006 (C.A. Santiviago), and DOD grant W81XWH-08-1-0719 (M. Zhao).

Accession numbers: Microarray hybridization data are accessible as GSE19609 at the GEO depository of the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov/geo/).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.