Comprehensive histological imaging of native microbiota in human glioma

Mounting evidence suggests that distinct microbial communities reside in tumors and play important roles in tumor physiology. Recently, a previous study profiled the composition and localization of intratumoral bacteria using 16S ribosomal DNA (rDNA) sequencing and histological visualization methods across seven tumor types, including human glioblastoma. However, their results based on traditional histological examinations should be further validated considering potential sources of contamination originating from sample collection and processing. Here, we aim to propose a three‐dimensional (3D) in situ intratumoral microbiota visualization and quantification protocol avoiding surface contamination and provide a comprehensive histological investigation on local bacteria within human glioma samples. We develop a 3D quantitative in situ intratumoral microbiota imaging strategy, combining tissue clearing, immunofluorescent labeling, optical sectioning microscopy, and image processing, to visualize bacterial lipopolysaccharide (LPS) within gliomas in a direct, contaminant‐free, and unambiguous manner. Through an automated statistical algorithm, reliable signals can be distinguished for further analysis of their sizes, distribution, and fluorescence intensities. In tandem, we also combined 2D images obtained from thin‐section histological methods, including immunohistochemistry and fluorescence in situ hybridization, to provide comprehensive histological imaging for local bacterial components within human glioma samples. We have, for the first time, achieved 3D quantitative imaging of bacterial LPS colonized in gliomas in a contamination‐free manner within human glioma samples. We also built the multiple histological evidence chain demonstrating the irregular shapes and sparse distribution of bacterial components within human glioma samples, mostly localized near nuclear membranes or in the intercellular space. This study provides favorable evidence for the presence of microbiota in human gliomas and provides information on the feature and distribution of bacterial components. The results, along with the integrated 3D quantitative intratumoral microbiota imaging method, are promising to provide insightful information into the direct interactions between the microbial community and the host in the tumor microenvironment.

Presidential Foundation of Zhujiang Hospital of Southern Medical University; Special Funds for the Cultivation of Guangdong College Students' Scientific and Technological Innovation microbiota imaging strategy, combining tissue clearing, immunofluorescent labeling, optical sectioning microscopy, and image processing, to visualize bacterial lipopolysaccharide (LPS) within gliomas in a direct, contaminant-free, and unambiguous manner. Through an automated statistical algorithm, reliable signals can be distinguished for further analysis of their sizes, distribution, and fluorescence intensities. In tandem, we also combined 2D images obtained from thin-section histological methods, including immunohistochemistry and fluorescence in situ hybridization, to provide comprehensive histological imaging for local bacterial components within human glioma samples. We have, for the first time, achieved 3D quantitative imaging of bacterial LPS colonized in gliomas in a contamination-free manner within human glioma samples. We also built the multiple histological evidence chain demonstrating the irregular shapes and sparse distribution of bacterial components within human glioma samples, mostly localized near nuclear membranes or in the intercellular space. This study provides favorable evidence for the presence of microbiota in human gliomas and provides information on the feature and distribution of bacterial components. The results, along with the integrated 3D quantitative intratumoral microbiota imaging method, are promising to provide insightful information into the direct interactions between the microbial community and the host in the tumor microenvironment.

K E Y W O R D S
fluorescent labeling, glioma, image processing, microbiota, three-dimensional visualization, tissue clearing 1 | INTRODUCTION Ever-increasing evidence has shown that the native microbiota constitutes an essential component of the tumor microenvironment across many tumor types [1][2][3]. The populations of bacteria colonized within tumors have been demonstrated to be tumor-type specific, which may directly regulate cancer initiation, progression, and patients' responses to therapies [1,3,4]. Glioma is the most common primary brain cancer and glioblastoma (GBM) is the most malignant kind with a poor prognosis and remains incurable [5]. Recently, Nejman et al. [4] verified that bacteria existed within seven tumors, including brain tumors, via combinational methods of immunohistochemistry, fluorescence in situ hybridization (FISH), electron microscopy, culturomics, and genomic sequencing. Nonetheless, their conclusions, despite their strict protocol of DNA sequencing against contamination, need to be further validated in consideration of potential contamination that can be traced back to sample resources and experimental procedures, especially when histological methods are involved. Meanwhile, these methods may lead to misinterpretation of the quantification and biogeography of microbial communities due to the lack of spatial resolution (e.g., thin sections) or individual-cell information (e.g., bulk sequencing for cell population).
To detect the presence, localization, and morphology of intratumoral bacteria, histological methods relied on ultrathin or thin tissue sections play a pivotal role, whereas often fail to avoid possible contamination on the surface of tissue sections and greatly limit the information capacity. Regarding these defects, the novel tissue clearing techniques are expected to provide a contamination-free manner of microbial detection for tumor samples via direct interrogation of intact tissues [6][7][8]. Tissue clearing and bacteria labeling approaches have been applied for in situ three-dimensional (3D) microbiota imaging, such as microbial identification after PACT (MiPACT) [9], tissue clearing-based, Damino acid labeling-facilitated (TiDaL) strategy [10], and progression in the 3D measurement of tuberculosis infections within lung tissues [11,12]. We previously developed Optical Properties-adjusting Tissue-Clearing (OPTIClear) [13], a tissue clearing method for fresh and archival human brain tissues, and later optimized it to Accurate delipidation with Optimal Clearing (Accu-OptiClearing) [8], which minimizes tissue distortion and destruction while achieving comparable transparency. We propose that this approach can be employed to clear brain tumor tissues, integrating with fluorescence imaging techniques and computational analysis methods, to visualize and quantify bacteria deep within tissues. By providing the quantitative in situ 3D information of intratumoral microbiota in single-cell resolution, the tissue clearing-based visualization approach will promise to help validate the presence of residing microbial communities within the tumor and interpret their role from a system biological perspective.
Here we undertook a comprehensive study on bacterial components within human gliomas combining 3D visualization of intact tissues and traditional histological staining of formalin-fixed, paraffin-embedded (FFPE) slices. We provided the first 3D, quantitative, and contamination-free information of bacterial LPS signals within human gliomas via the proposed tissue clearing-based intratumoral microbiota imaging protocol. Combining these pieces of histological evidence, we hope to give support to the presence of bacteria in gliomas and contribute to a comprehensive analysis in regards to their sizes, morphologies, and spatial distributions. Incorporating more methodologic improvements, this 3D in situ quantitative intratumoral microbiota imaging pipeline is promising to reveal the panorama of the human glioma microbiota and is expected to provide insightful information into the direct host-microbiota interactions in the glioma microenvironment soon.

| Human glioma samples
Three human glioma samples used in this study were randomly selected and obtained during neurosurgeries at the Zhujiang Hospital (Table S1). Informed consent has been obtained for utilizing the resected tissues for research, with prior approval by the Medical Ethics Committee of Zhujiang Hospital of Southern Medical University (Approval Number: 2018-SJWK-004 and 2020-YBK-001-02). The specimens were resected and immediately fixed in neutral buffered formalin (NBF) and sent for diagnostic pathological examination. Tissues not used in the clinical-pathological examination were salvaged for this study. These consisted of fragmented pieces of brain tumors, which have not been embedded in paraffin wax. The total duration of tissue fixation in NBF was about 9-11 months at 4 C. The three samples were diagnosed by clinicians and graded according to WHO classification.

| Animals
C57BL/6 mice (8-9 weeks old, 18-22 g, male) were obtained and raised in the Experimental Animal Centre of Zhujiang Hospital of Southern Medical University, Zhujiang Hospital, and fed in a specific pathogen-free lab with constant temperature and humidity. The cage, pad, feed, etc., were sterilized by high-pressure steam and replaced regularly. All animal experiments in this study were performed in strict compliance with the ethical principles of experimental animal welfare.

| Mouse dissection and organ collection
A C57BL/6 mouse was deeply anesthetized with 1% pentobarbital sodium. The mouse was perfused with saline, followed by 4% (wt/vol) paraformaldehyde (PFA) fixation. The total intestinal tract was dissected separately and rinsed with 4% PFA to remove intestinal contents. The sample was post-fixed in 4% PFA for 2 days at 4 C. Another C57BL/6 mouse was killed by neck-breaking and the brain was dissected and immediately collected. The brains were post-fixed with 4% PFA at 4 C for 3 days. Before tissue clearing, tissues were gently rinsed with 0.01% (wt/vol) phosphate-buffer saline (1 Â PBS) twice.

| Bacteria smear and immunolabeling
The preparation of bacteria smears and immunolabeling have followed the literature [14]. A loopful of gram-negative bacteria (Escherichia coli strain ATCC25922) cultured in a Columbia agar base (Guangzhou Dijing Microbial Technology Co., Ltd. #LS0109) were transferred to clean EP tubes and fixed in PFA for 1 day. EP tubes were centrifuged (15 min, 4000-8000g, at 4 C) and the supernatant was removed. The bacteria pellet was resuspended in PBS and washed three times. A loopful of the resuspended bacteria was transferred into a drop of ddH 2 O in the center of clean slides to make a suspension, air-dried. The smears were fixed in 95% (vol/vol) ethanol for 1 min at room temperature (RT) and dipped in PBS, air-dried. Immunolabeling of bacteria was performed according to the standard staining method. Primary antibodies (Lipopolysaccharide Core, mAb WN1 222-5, HycultBiotech #HM6011, 1:100 dilution) were applied on smears for 30 min at 37 C and secondary antibodies (Donkey anti-Mouse IgG [H + L] Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor Plus 594, Thermo Fisher Scientific #A32744, 1:100 dilution) were added for 30 minutes at 37 C. The slides were rinsed in PBS for 30 min at RT and then blotted with Kimwipe paper. The slides were mounted with mounting medium with 4 0 ,6diamidino-2-phenylindole (DAPI) (Fluoroshield Mounting Medium With DAPI, ABCAM #ab104139) and coverslips to the smears. The slides were observed with confocal laser scanning microscopy (LSM 800, Carl Zeiss, Germany) equipped with the objectives Plan-Apochromat 63X/1.40 Oil DIC M27.

| Multiphoton laser scanning microscopy (MPLSM) imaging
All cleared tissues were mounted on 60-mm cell and tissue culture dishes wetted with OPTIClear solution (about 200 μL) under the microscopy. Images were obtained with an MPLSM (Olympus, FVMPE-RS, Tokyo, Japan) equipped with XLPLN10XSVMP (Â10/0.6 NA) objective lens. We determined the optimal excitation wavelength of fluorescent dyes by gradually adjusting the wavelength of excitation light from 680 to 1050 nm. The excitation laser wavelengths were adjusted to 750 and 900 nm for DAPI and Alexa Fluor 594, respectively. The data were reconstructed and analyzed with Imaris, version 9.0.2 (Bitplane AG, Zurich, Switzerland).

| Image processing
Laser power and gain values were adjusted to the optimum for each image so that the fluorescence of positive signals and cell nucleus can be displayed clearly. 3D image reconstruction was made with Imaris imaging software (version 9.0.1, Bitplane AG, Zurich, Switzerland). Subsequently, each fluorescence image was processed by MATLAB (version R2019b) for counting the objects and calculating the mean fluorescence intensity and volume of each bacterial LPS fluorescent signal. In this process, we first binarized each slide along the depth in the Z-axis of one 3D image by a specific threshold, which was obtained based on the mean gray value of the slide, to roughly segment the bacterial signals. To further refine the segmentation, we conducted a region growing method for each object to make it include the surrounding pixels with high and close gray values. All the slides of one image were rearranged to screen the real bacterial objects with a diameter of 0.5-5 μm. The final statistics of bacteria loads, objective sizes, and the mean fluorescence intensities were based on those screened bacterial LPS fluorescent signals. The MATLAB code is available at https://github.com/PRBioimages/Fluorescence-objectcounting. It can be used for high-throughput automated processing of 3D fluorescence images.

| 3D quantitative imaging of bacterial LPS fluorescent signals within human glioma samples
To eliminate the negative impacts of potential contamination during the tissue sampling and handling of tumor tissues for intratumor microbiota detection, we developed an Accu-Opticlear-based tissue clearing protocol [8] to observe microbes within tissues, and for the first time accomplished 3D visualization of bacterial LPS fluorescent signals in human glioma samples ( Figure 1A-C). We used antibodies to target the LPS cores anchored in the cell wall of Gram-negative bacteria and verified their specificity ( Figure S1), which also had been well validated by Nejman et al. [4] examining human tumor tissue microarrays over seven tumor types. Here, the human glioma samples were sliced into 500 μm thick and cleared by the Accu-Opticlear protocol [8] in combination with an autofluorescence bleaching step [15] in clean environments with sterilized reagents and equipment. The most superficial parts, around 50-100 μm, were ignored during imaging by MPLSM, and an internal 100 μm part was scanned at a light-cutting interval of 1 μm (Figure 1A-C).
The 3D reconstruction images and videos clearly showed the sporadic distribution and irregular shapes of LPS fluorescent signals, which were mostly located near the nuclear membranes or diffused in intercellular spaces ( Figure 1D; Movies S1 and S2). We provided an automated image processing pipeline to exclude fluorescent signals with improper sizes and allow quantitative analysis in terms of the load, size, and fluorescent intensity of the signals ( Figure 1E). Quantitative analysis also suggested that these signals were randomly localized and of uneven sizes, with an average diameter of 2.17 ± 0.80 μm. To improve the quality of imaging, we also performed different autofluorescence quenching methods (with CuSO 4 treatment ( Figure 2B) and with SBB treatment (Figure 2C)), in comparison with that without autofluorescence quenching step (Figure 2A), to better characterize the size and 3D distribution of the bacterial LPS signals within glioma tissues. Whereas the low sample amount and microscopic fields in this study limited a thorough investigation of the universal features and distribution patterns of bacterial LPS in gliomas. In addition, these LPS fluorescent signals within glioma exhibited neither the typical features nor complete profiles of Gram-negative bacteria, which possibly attributes to the deficiency of cell walls and envelope transformation of intracellular bacteria [4]. Also, steps such as formalin fixation and permeation can alter bacterial morphology and introduce artifacts [16] ( Figure S1).
Taken together, despite further requirements for methodological improvements, this 3D quantitative in situ intratumoral bacteria imaging method provides the first direct and contaminant-free image of bacterial LPS fluorescent signals within human glioma samples. We also developed a customized statistical algorithm to accurately capture LPS fluorescent signals and give a quantitative description of their morphology and distribution. This integrated protocol for detecting and analyzing intratumoral bacterial components is also applicable to the investigation of diverse tumors and promises to advance the direct study of host-microbe interactions.

| Bacterial LPS and RNA can be visualized in human glioma samples
To further complement the acquired 3D information, we performed traditional histopathological examinations for tissues from the same samples. We adopted antibodies against bacterial LPS and lipoteichoic acid (LTA) to target Gram-negative and Gram-positive bacteria via immunohistochemistry staining, respectively. We demonstrated similar results as Nejman et al. [4] reported, that LPS was detected in glioma samples while LTA was absent (Figure 3). Universal 16S rRNA FISH probes were applied to detect bacteria, with antisense probes serving as control. Distinct localization of bacterial 16S rRNA signals was found mostly alongside the nuclear membrane, irregular in shape (Figure 3). Quantification of fluorescent signals showed sizes ranging from 0.26 to 4.26 μm, with an average diameter of 0.59-1.85 μm. The results of 16S rRNA FISH staining combined with those of bacterial LPS immunostaining further indicated the atypical and variable morphologies of these bacterial components in tumor tissues. Moreover, in consecutive slices (4 μm thick), the LPS and 16S rRNA staining showed close localization, which is similar to the results published by Nejman [4] (Figure S2). Although the 2D

| DISCUSSION
In the present work, we combined multiple histological methods to investigate bacteria in human glioma samples. We developed an Accu-OptiClearing-based contaminant-free 3D pathology protocol to image bacterial components in human glioma, allowing further analysis for their quantification and distribution characteristics. This novel attempt, deriving from our initial work for clearing human brain tissues [8,13], contributed to previous advances in visualizing microbes in their native context in hosts [9,10,12] and expanded the types of tissue applicable. This 3D histology imaging protocol is optimized by (a) collecting, sectioning, and processing the sample in clean environments and using sterilized reagents and equipment; (b) including negative and positive controls to monitor contamination and determine the effectiveness of the results; (c) studying the inside rather than the surface of intact cleared samples; (d) quenching autofluorescence to eliminate interference from unwanted fluorescent signals. We also represented, to our best knowledge, the first attempt to visualize and measure the bacterial LPS fluorescent signals in situ in the context of 3D histology free of contamination. This protocol also gives wide access to the visualization and quantification of bacterial signals within diverse tumors, aiming to provide a systematic characterization of the distribution and morphologies of intratumoral bacteria in situ. Besides, we built the multievidence chain that combines 2D and 3D histology to comprehensively investigate these signals within human glioma, which will help verify the existence of intratumoral bacteria in human glioma.
Although we have optimized the 3D imaging protocol to exclude potential contamination, there remain possible impacts of post-mortem microbial translocation and other unaware contaminants, a universal problem in examine post-mortem samples. Procedures such as shortening post-mortem intervals, including sterile operations, and coupling with negative controls can help reduce errors [17]. Better approaches may be to identify microbes in fresh tumor tissues by vitro culture method and even directly track microbes in vivo. Immunofluorescent labeling for thick tissues raises recurring difficulties in eradicating non-specific fluorescence from non-specific binding of antibodies and autofluorescence derived to natural pigments (e.g., lipofuscin, flavin, mitochondria, hemoglobin, etc.) or formalin fixation. Autofluorescence in biological tissues, unfortunately, exhibits broad spectral ranges of excitation and emission wavelengths and approximate features similar to bacterial fluorescence profiles [18][19][20]. A preference for long-wavelengths fluorophores is recommended since the typical autofluorescence is mostly overlapped with green-blue fluorophores. Studies report that CuSO 4 and SBB stains unwanted autofluorescence black, such as lipofuscin, an aging-related pigment that accumulates in many cells including those of the central nervous system, via boundary surface adsorption [19,21]. To minimize the interference of autofluorescence, here we used chemicals (CuSO 4 and SBB) [21] to mask autofluorescence while enabling the maintenance of tissue transparency and specific fluorescent labels. Facilitated by computing algorithms, fluorescence thresholding can also be automatically performed. However, both approaches have drawbacks, such as chemical quenching may reduce immunolabeling and the accuracy of automated spectral thresholding is unstable [21]. The protocol also needs to be reassessed whenever an experiment element is adjusted, to produce the most biologically relevant results. Problems also arise in the accuracy and sensitivity of 3D quantitative analysis for the signals, considering tissue deformation (e.g., shrinkage and expansion) and immunolabeling inaccessibility to all the targets during tissue clearing, which may ultimately result in inaccurate parameters (e.g., diameter and quantity, etc.) [12]. Meanwhile, the confined sample size and microscopic fields of this study reduced the validity of the analysis. As a preliminary study, the incomplete nature of this work limited the information we acquired and prevented us from interpreting the constitution and roles of intratumoral bacteria. We hope to increase the reliability of fluorescent labeling for bacteria in tissue sections and 3D quantitative analysis in future research, and explore the underlying molecular mechanisms. This work also highlights the dilemma in histological detection of intratumoral microbiota, where results from multiple methods should be combined to make a comprehensive judgment.
While the present study adds to the picture of the glioma microbiota, more issues warrant further investigation. The tumor-specific composition and function of the tumor microbiome have been the upsurge of omics sequencing and clustering analysis [22][23][24]. For the brain, the organ thought to be distinctively immuneprivileged against microbial invasion by physiological barriers (e.g., blood-brain barrier [BBB]), how microbiota evolves and resides in brain tumors post an interesting question. Roberts et al. [25] once reported finding rod-shaped bacteria in healthy human postmortem brains by EM and proposed that bacteria may enter the brain through the BBB or via nerves innervating the gut, however, lacking further verification. Although the bidirectional "gut-brain axis" has been defined to describe the interaction of the gut microbiome and the brain, only indirect pathways have been confirmed so far [3,[26][27][28]. While most studies on the direct host-bacteria interaction focused on how local microbial communities affect colonized tumors, yet remaining little is known [29]. Driven by the advances in bacterial probing and characterization, such as the development of STAMP [30] and HIPR-FISH [31] techniques, tissue clearing technology is expected to be armed as a powerful tool in profiling the human tumor microbiota. We expect that 3D in situ quantitative imaging of intratumoral microbiota in their native context with single-cell resolution will promote the dissection of the intricate interactive network among microbiota, tumor cells, immune cells, and other components in the tumor microenvironment. design of the experiments. Dian He, Haitao Sun, Hei Ming Lai, and Yunhao Luo developed the methodological approaches. Haitao Sun, Hongbo Guo, and Hongwei Zhou provided the lab resources; Haitao Sun, Wutian Wu, Dian He, and Jiajia Zhao interpreted the data; Jiajia Zhao, Haitao Sun, and Dian He drafted the manuscript. Yunhao Luo, Yiquan Ke, and Linlang Guo acquired the patient samples and information. Dian He, Jiajia Zhao, Ting Li, Yunhao Luo, and Jianhao Liang performed the assay and acquired the data. Jiajia Zhao, Haitao Sun, Dian He, Wutian Wu, Yingying Xu, and Xiaodu Yang analyzed and curated the data. Haitao Sun, Wutian Wu, Hei Ming Lai, and Hongwei Zhou revised the manuscript. All authors read and approved the final manuscript.

ETHICS STATEMENT
The work was conducted under ethical approval held by the Clinical Biobank Centre at Zhujiang Hospital of Southern Medical University (approval number 2020-YBK-001-02). Informed consent has been obtained for utilizing the resected tissue for research, with prior approval by the Medical Ethics Committee of Zhujiang Hospital of Southern Medical University (approval number 2018-SJWK-004).

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.