Nanoscale Bacteria‐Enabled Autonomous Drug Delivery System (NanoBEADS) Enhances Intratumoral Transport of Nanomedicine

Abstract Cancer drug delivery remains a formidable challenge due to systemic toxicity and inadequate extravascular transport of nanotherapeutics to cells distal from blood vessels. It is hypothesized that, in absence of an external driving force, the Salmonella enterica serovar Typhimurium could be exploited for autonomous targeted delivery of nanotherapeutics to currently unreachable sites. To test the hypothesis, a nanoscale bacteria‐enabled autonomous drug delivery system (NanoBEADS) is developed in which the functional capabilities of the tumor‐targeting S. Typhimurium VNP20009 are interfaced with poly(lactic‐co‐glycolic acid) nanoparticles. The impact of nanoparticle conjugation is evaluated on NanoBEADS' invasion of cancer cells and intratumoral transport in 3D tumor spheroids in vitro, and biodistribution in a mammary tumor model in vivo. It is found that intercellular (between cells) self‐replication and translocation are the dominant mechanisms of bacteria intratumoral penetration and that nanoparticle conjugation does not impede bacteria's intratumoral transport performance. Through the development of new transport metrics, it is demonstrated that NanoBEADS enhance nanoparticle retention and distribution in solid tumors by up to a remarkable 100‐fold without requiring any externally applied driving force or control input. Such autonomous biohybrid systems could unlock a powerful new paradigm in cancer treatment by improving the therapeutic index of chemotherapeutic drugs and minimizing systemic side effects.

S-2 Figure S2. NanoBEADS Viability. Composite bright-field and fluorescence microscopy images of unmodified bacteria and NanoBEADS (left column). The fluorescence microscopy images (right column) show live bacteria (green) and dead bacteria (red). Conjugation of nanoparticles (22±14) to bacteria through biotin-streptavidin affinity bonds does not affect bacteria viability. All scale bars are 20 µm.

I. Supplementary Figures and Tables
S-3 Figure S4. Effect of Antibody Concentration on NanoBEADS Formation. At nanoparticle to bacteria assembly ratio of 50:1, the fraction of bacteria with at least one nanoparticle attached steadily rose with increasing concentration of antibody till the maximum concentration of 10 µg mL -1 .

Figure S3. Cancer Cell Viability in Presence of TIPS-pentacene-loaded PLGA Nanoparticles.
Composite fluorescence microscopy images show live cancer cells (blue) and dead cancer cells (green) in absence (control) and presence of 6×10 8 mL -1 TIPS-pentacene-loaded PLGA Nanoparticles (red) after 12 hours of incubation. No dead cells were observed, suggesting that the nanoparticles do not affect cancer cell viability. All scale bars are 50 µm.

Figure S7. Radial distribution of Therapeutic Agents in Multicellular Tumor Spheroids. Absolute (A)
and normalized (B) therapeutic agent numbers vs. radial location, measured from the surface of the tumors. For HCT-116 tumor spheroids, N=10 for PEGylated bacteria, N=9 for NanoBEADS, and N=7 for nanoparticles. For U87MG tumor spheroids, N=7 for PEGylated bacteria, N=7 for NanoBEADS, and N=7 for nanoparticles. For 4T1 tumor spheroids, N=7 for PEGylated bacteria, N=4 for NanoBEADS, and N=3 for nanoparticles.    were graded on a scale from 0-4 on the percentage of the sections affected by necrosis, EMH, and inflammation. The individual scores were then summed to create a composite score. There were no significant differences between the groups with regards to spleen scores. The arrows are pointing to normal lymphoid tissue in the spleen and the asterisks are pointing to the pale EMH in the adjacent red pulp. All scale bars are 200 µm.

Figure S12. Intratumoral Transport of Therapeutic Agents in a Murine Breast Cancer Model.
Confocal microscopy images showing the penetration of PLGA nanoparticles, PEGylated S. Typhimurium VNP20009, and NanoBEADS through 4T1 mammary tumors in BALB/c mice 48 h after intratumoral injection. The point of injection was considered to be at the end of the needle track, marked with asterisks on the images (top row). Zoomed-in micrographs with white arrows pointing to representative therapeutic agents (bottom row). S-9

II. Image Processing for Quantitation of the Spatial Distribution of Therapeutic Agents in Tumor Slices
We developed a custom image processing routine to quantitate and produce a 3D map of the therapeutic agents in tumor slices that were imaged using a laser scanning confocal microscope. Our algorithm is comprised of three components: (i) Calibration of the size and intensity of the object of interest: sample images containing sparsely distributed singular therapeutic agents were processed to define average object size or the number of pixels that represent a therapeutic agent. Next, an analytical function was fit to the distribution of fluorescence intensity along the z-axis for a series of representative z-stacked images, and fluorescent signal dissipation with respect to distance in the z-dimension was determined. Together, this data provided the object size information needed to accurately quantify the number of objects; (ii) Conversion of the grayscale fluorescent intensity image into a binary image: User-defined thresholding parameters including global and local gray-level intensity values were applied to convert grayscale input images into binary images; (iii) Construction of the 3D distribution map: using binary images produced in (ii) and the calibration data from (i), a compensation algorithm was utilized to enable the enumeration of fluorescent therapeutic agents in 40 µm-thick tumor slices.

II. A. Error Analysis
In order to determine the accuracy of our image processing routine, we compared the number of fluorescent objects detected by the image processing routine with the number of objects detected by manual counting. The error was calculated according to: where N manual is the number of objects determined by manual counting and N automated is the number of objects reported by the image processing routine. In the case of nanoparticles, the manual counting was carried out using scanning electron microscopy (SEM) images, whereas, for the larger bacteria and NanoBEADS, the manual counting was done from the confocal microscopy images. It should be noted that, given the co-localization of the nanoparticles and bacteria in the case of NanoBEADS (Figure S6), only the signal from bacteria (visualized at 543 nm excitation wavelength with a 553-624 band-pass filter) was S-10 Figure S14. Image Processing Error Analysis. The error in quantitation of the spatial distribution of therapeutic agents was determined by comparing the number of fluorescent objects reported by our image processing routine with the number of fluorescent agents determined through manual counting. The mean error was found to be 7.8 ± 1.9 % in the case of bacteria and NanoBEADS and 13.1 ± 1.9 % in the case of nanoparticles.
used for image processing to avoid any error due to variation in the number of PLGA nanoparticles (visualized at 633 nm excitation wavelength with a 635-750 band-pass filter). The bacterial nanoparticle load information was separately determined through analysis of SEM images (Figure 1). As shown in Figure S14, we found that the mean error in the bacteria and NanoBEADS case was 7.8 ± 1.9 % (n = 9). The error largely stemmed from the variability in bacteria orientation which leads to variation in their projected area. In the case of nanoparticles, the mean error was 13.1 ± 1.9 % (n = 22). The larger error was due to the smaller size of the nanoparticles combined with the variability in the 3D clustering of the nanoparticles.