Correlative light electron ion microscopy reveals in vivo localisation of bedaquiline in Mycobacterium tuberculosis–infected lungs

Correlative light, electron, and ion microscopy (CLEIM) offers huge potential to track the intracellular fate of antibiotics, with organelle-level resolution. However, a correlative approach that enables subcellular antibiotic visualisation in pathogen-infected tissue is lacking. Here, we developed correlative light, electron, and ion microscopy in tissue (CLEIMiT) and used it to identify the cell type–specific accumulation of an antibiotic in lung lesions of mice infected with Mycobacterium tuberculosis. Using CLEIMiT, we found that the anti-tuberculosis (TB) drug bedaquiline (BDQ) is localised not only in foamy macrophages in the lungs during infection but also accumulate in polymorphonuclear (PMN) cells.


Introduction 28
An effective chemotherapy against bacterial infections must include antibiotics with 29 pharmacokinetic properties that together allow penetration into all infected 30 microenvironments [1]. Antimicrobial penetration is especially important for the treatment of 31 infections where antibiotics need to reach intracellular bacteria [2], including Mycobacterium 32 tuberculosis. In tuberculosis, treatment requires at least three antibiotics for six months [3], and 33 we do not fully understand why this extended treatment is needed. In this context, 34 understanding how tissue environments affect antibiotic localisation, exposure, and 35 consequently efficacy against the pathogen, is crucial [4]. 36 37 Although it is critical to define if antimicrobials are able to reach their intracellular targets, 38 imaging of antibiotics (and drugs in general) at the subcellular level in infected tissues remains 39 challenging. Only recently have studies in vivo determined antibiotic distributions in 40 granulomatous lesions by matrix-assisted laser desorption-ionisation mass spectrometric 41 imaging (MALDI-MSI) [5]. However, this approach only allows analysis at the tissue level and 42 does not reach subcellular or even cellular resolution [6]. On the other hand, nanoscale 43 secondary ion mass spectrometry (nanoSIMS) has been used to visualise drugs at 50 nm 44 resolution in cells [7] and tissues [8]. However, there are limitations with this method such as 45 the lack of correlation with other available imaging modalities that provide spatial information 46 of specific cell types localisation and function. Thus, correlative approaches are needed to 47 obtain both spatial localisation of drugs and biologically relevant information from 48 experimental systems [9]. Recently, a correlative imaging approach combining light, electron 49 and ion microscopy (CLEIM) has been developed for subcellular antibiotic visualisation in 50 vitro cultured cells [10]. However, there are currently no approaches available that allow 51 correlative studies at the subcellular resolution in vivo.

53
Results 54 55 With the aim to define the subcellular localisation of antibiotics in infected cells within tissues, 56 we used a mouse model of tuberculosis. Our goal was to develop an imaging approach to 57 analyse the distribution of antibiotics from complex tissues to individual cells at the subcellular 58 level in infected lungs. For that, we infected susceptible C3HeB/FeJ mice with Mycobacterium 59 tuberculosis H37Rv expressing fluorescent E2-Crimson via aerosol infection ( Figure 1A). The 60 C3HeB/FeJ susceptible mouse strain develops necrotic lesions in the lung that better 61 recapitulate human granulomas, a hallmark of tuberculosis infection [11]. After 21 days of 62 infection, mice were treated daily for five days either with control vehicle or 25 mg/kg of the 63 anti-mycobacterial antibiotic bedaquiline (BDQ). As previously reported [12], this treatment 64 reduced approximately ten-fold the bacterial loads in the lungs, as measured by colony forming 65 units (CFU) counting (Figure S1A/B). Following treatment, mice were euthanised and fixed 66 by perfusion with formalin. Lungs were removed and granulomatous lesions were visualised 67 by Micro Computed Tomography (µCT, Figure 1A, Figure S1C and Movie S1). Replicate 68 lung tissues were embedded in agarose for further processing and imaging. 69 70 One of the main technical challenges of our attempt to define if the antibiotic reached 71 intracellular bacteria was to identify and correlate across the different imaging modalities and 72 scales the infected cells present in the lung. We devised a strategy that included the 73 identification of a granulomatous lesion within 100 µm thickness sections and non-destructive 74 3D imaging by confocal laser scanning microscopy of the entire section as well as the region 75 of interest ( Figure 1B and 1C). Vibratome sections were stained with DAPI to visualise nuclei 76 and BODIPY 493/503 to visualise lipid droplets (LD), previously shown to accumulate in 77 foamy macrophages in necrotic lesions [13]. In agreement with previous studies, we found that 78 granulomatous lesions were heavily enriched in LD-laden foamy macrophages ( Figure 1B and 79 1C). After fluorescence imaging, sections were recovered and resin embedded for electron 80 microscopy. In order to correlate the 3D fluorescence microscopy with the electron 81 microscopy, sections were analysed by µCT 3D scanning ( Figure S2). This approach enabled 82 the precise localisation of the ROI previously imaged by fluorescence, and the angle correction 83 during sectioning ( Figure S2). In this way, the section obtained for Scanning Electron 84 Microscopy (SEM) and nanoSIMS could be matched to the 3D fluorescence image with a high 85 degree of accuracy ( Figure S2). The sections were then imaged by SEM ( Figure 1D) and the 86 same section was then coated with 5 nm gold and transferred for nanoSIMS analysis. BDQ 87 contains a bromine atom, so we determined its localisation by the intensity of the 79 Br ion 88 signal [10]. The regions imaged by SEM were identified using the optical microscope in the 89 nanoSIMS. The sample was scanned with a focused 133 Cs + and secondary ions ( 12 C -, 12 C 14 N -, 90 79 Br -, 32 Sand 31 P -) and secondary electrons were collected ( Figure 1E). The 12 C 14 Nand 31 P -91 signals were useful to show the morphology of cells and tissues, with 12 C 14 Nsignals largely 92 from proteins and the highest 31 Psignals are from nucleic acids and structures we believe are 93 polyphosphates in Mtb. 94 To correlate across imaging modalities with subcellular resolution, endogenous structures were 95 used as landmarks. LD were located by fluorescent staining in the optical image, ultrastructure 96 in the SEM image, and 32 Ssignal in the ion image. The 32 Ssignal was due to the 97 osmium/thiocarbohydrazide staining of lipids. Bacteria were localized by fluorescence (E2-98 Crimson signal), ultrastructure and 31 Psignal in the ion image. The cell nucleus was aligned 99 using ultrastructure and the 31 Psignal ( Figure 1F). Concurrent with previous CLEIM in vitro 100 studies, we found that BDQ accumulated heterogeneously in LD and Mtb, with particularly 101 high levels in infected foamy macrophages ( Figure 1F and Figure S3). Importantly, some 102 bacteria contained high levels of the antibiotic whereas others did not show any signal, 103 indicating that the antibiotic is not able to evenly reach throughout intracellular bacteria present 104 in the infected tissue ( Figure 1F). 105 106 Taking advantage of this method, we then focused on a more quantitative approach ( Figure  107 S4) to analyse intracellular antibiotic localisation in the lung lesions. For that, we performed a 108 combined tile scanning by SEM and nanoSIMS covering larger areas of the tissue (Figure 2A).

109
This allowed to define the distribution of BDQ in single cells and bacteria (Movie S2). 110 Unexpectedly, we found that BDQ not only localised in lipophilic environments (e.g. in LD) 111 but also in non-lipophilic cellular environments. Specifically, we found that BDQ strongly 112 accumulated in polymorphonuclear cells (PMN). Antibiotic-rich PMN were present both 113 alongside ( Figure 2B-C) and away from areas enriched with foamy macrophages ( Figure 2D).

114
In contrast to macrophages where the 79 Brsignal was primarily associated with LD; in PMN, 115 the 79 Brsignal was not only associated with granules but also with the cytosol (Figure S5).

116
Thus, in tissues, BDQ accumulates in a cell-type dependent manner across two cell populations 117 (foamy macrophages and PMN) with very different metabolic and functional properties. 118 Confirming our previous observations, quantitative analysis revealed that BDQ 119 heterogeneously accumulated in LD and Mtb ( Figure 2E). Both Mtb outside and inside LD 120 accumulated BDQ ( Figure 2E). These PMN cells are likely neutrophils recruited to the 121 granuloma as reported in this mouse model of TB infection [14]. Neutrophils are rich in 122 Myeloperoxidase (MPO), a peroxidase that produces hypochlorous acid from hydrogen 123 peroxide and chloride anion or hypobromous acid if bromide anion is present [15]. However, 124 in untreated mice the 79 Brsignal was significantly lower and only slightly associated with 125 PMN granules, indicating that the 79 Brsignal was primarily coming from the antibiotic 126 ( Figure S5). 79 Br -/ 12 C 14 Nwere used when comparing the BDQ-treated and non-treated 127 tissues. The normalisation to 12 C 14 Nwas to compensate possible minor variations in the 128 primary ion current during imaging. 129 130 Discussion 131 Altogether, we report the development of a correlative approach in tissue to define the 132 subcellular localisation of antibiotics in infected cells within tissues. This multimodal imaging 133 approach represents a powerful methodological advance to investigate if drugs reach their 134 intracellular targets. Using this approach, we identified that in the lungs of M. tuberculosis 135 infected mice, the antibiotic BDQ heterogeneously localised to intracellular bacteria and LD 136 of foamy macrophages. We also found that BDQ significantly accumulated in specific cell 137 types such as PMN, likely neutrophils, recruited into granulomatous lesions. Therefore, 138 CLEIMiT enabled us to characterise the antibiotic distribution across multiple cell types, 139 revealing multiple other niches of drug accumulation. 140 141 CLEIMiT is readily applicable to other drugs and not only for antibiotics or bromine-142 containing drugs but also any drugs detectable by nanoSIMS. Ion microscopy methods using 143 nanoSIMS represents a good combination of spatial resolution and sensitivity to map drugs 144 that contain elements other than Bromine that are low in the biological systems such as 145 platinum [ via oral gavage while the control group were given only 2-hydroxypropyl-β-cyclodextrin. At 189 the end of treatment, mice were euthanised by anaesthesia, then the lungs were either perfused 190 with 10% neutral buffered formalin and excised ( Figure  were then set in 4% LMA and imaged using a Xradia 510 Versa 3D X-ray microscopes (Zeiss,204 Germany) with the following acquisition setting: 0.4X objective, pixel size = 7 µm, pixel 205 binning of 2, source filter = LE1, Voltage = 40 kV, Wattage = 3.0 W. Tomogram reconstruction 206 was carried out using the Zeiss Scout and Scan Software (Zeiss, Germany). Visualisation and 207 fine measurements were taken from a 3D volume reconstruction using Zeiss XM3D viewer 208 software (Zeiss, Germany). 209 210 µCT imaging of resin embedded tissue: A tissue slice was imaged using a Xradia 510 Versa 211 3D X-ray microscopes (Zeiss, Germany) with the following acquisition settings: 4X objective, 212 pixel size = 2.8 µm, pixel binning of 2, source filter = LE2, Voltage = 40 kV, Wattage = 3.0 213 W. Tomogram reconstruction was carried out using the Zeiss Scout and Scan Software (Zeiss, 214 Germany). Visualisation and fine measurements were taken from a 3D volume reconstruction 215 using Zeiss' XM3D viewer software (Zeiss, Germany). 3D measurements of the resin section 216 were used to give precise co-ordinates for the location of the fluorescently imaged area in the 217 resin block ( Figure S2B) and to determine the precise angle of advance for the diamond knife 218 when trimming the resin block. 219 220 Vibratome Sections 221 The lungs were separated into the four constituent lobes of the right lung (superior, middle, 222 inferior cover glass (NA=1.5) was gently placed upon the tissue and the medium allowed to set. An 239 inverted Leica TCS SP8 microscope running LAS X acquisition software with Navigator 240 module (Leica Microsystems, Germany), equipped with 405 nm, Argon laser, 561 nm, 633 nm, 241 and HyD detectors was used to image the tissue fluorescence with the following Lasers: 405nm 242 (DAPI), 488nm (BODIPY) and 561nm (Mtb-E2Crimson). In the first instance, the entire tissue 243 section was imaged with a tile scan using the 10x objective lens. Regions of interest (ROI) 244 were then identified based upon areas of tissue showing high degrees of cellular infiltration, 245 indicated by DAPI staining, and the accumulation of highly-lipid foamy cells, indicated by 246 BODIPY staining. Selected ROI were then imaged at higher resolution using a 40x oil 247 objective and z-stack. Voxel size was adjusted to half the theoretical limit of the lens in x and 248 y and 0.5 µm in z. Fields of view were chosen to include cellular architecture such as airway 249 passages as well as erythrocytes and vessels of Referring to measurements from the 3D volume reconstruction, generated by µCT, the sample 278 block was trimmed, coarsely by a razor blade then finely trimmed using a 35 o ultrasonic, 279 oscillating diamond knife (DiATOME, Switzerland) set at a cutting speed of 0.6 mm/s, a 280 frequency set by automatic mode and a voltage of 6.0 V, on a ultramicrotome EM UC7 (Leica 281 Microsystems, Germany) to remove all excess resin and tissue surrounding the ROI. Precise 282 measurements, derived from the µCT reconstruction, were used to further cut into the tissue, 283 to the depth corresponding with the fluorescent area previously imaged. 284 285 Nanoscale secondary ion mass spectrometry (nanoSIMS) 286 The sections were imaged by SEM and nanoSIMS as previously described [10]. 500 nm 287 sections were cut using ultramicrotome EM UC7 (Leica Microsystems, Germany) and mounted 288 on 7 mm by 7 mm silicon wafers. Sections on silicon wafers were imaged using a FEI Verios 289 SEM (Thermo Fisher Scientific, USA) with a 1 kV beam with the current at 200 pA. The same 290 sections were then coated with 5 nm gold and transferred to a nanoSIMS 50L instrument 291 (CAMECA, France). The regions that were imaged by SEM were identified using the optical 292 microscope in the nanoSIMS. A focused 133 Cs + beam was used as the primary ion beam to 293 bombard the sample; secondary ions ( 12 C -, 12 C 14 N -, 79 Br -, 32 Sand 31 P -) and secondary 294 electrons were collected. A high primary beam current of ∼1.2 nA was used to scan the sections 295 to remove the gold coating and implant 133 Cs + to reach a dose of 1x10 17 ions/cm 2 at the steady 296 state of secondary ions collected. Identified regions of interest were imaged with a ~3.5 pA 297 beam current and a total dwell time of 10 ms/pixel. Scans of 512 × 512 pixels were obtained. 298 299 300

Image alignment 301
Tissue derived micrograph and nanoSIMS/micrograph correlation: ion and fluorescent images 302 were aligned to EM micrographs with Icy 2.0.3.0 software (Institut Pasteur, France), using the 303 ec-CLEM Version 1.0.1.5 plugin. No less than 10 independent fiducials were chosen per 304 alignment for 2D image registration. When the fiducial registration error was greater than the 305 predicted registration error, a non-rigid transformation (a nonlinear transformation based on 306 spline interpolation, after an initial rigid transformation) was applied as previously 307 described [24]. 308 309 Quantification and Statistical analysis 310 Ion quantification: secondary Ion signal intensities were quantified in ImageJ with the 311 OpenMIMS v3.0.5 plugin. 312 313 Quantification of BDQ within bacteria: bacteria (totaln = 472) were manually outlined with the 314 assistance of SEM images and the 31 Psignal. Ratio values ( 79 Br -/ 12 C 14 N -) for bacteria were 315 divided by the area of their respective ROI to give mean normalised pixel intensity in arbitrary 316 units (AU) for each condition. Mean normalised pixel intensity in arbitrary units (AU) per ROI 317 was plotted against condition using Graphpad Prism 8 software and two-tailed p-value was 318 determined by an unpaired, non-parametric Mann-Whitney U test to assess statistical 319 significance. 320 321 Quantification of BDQ in lipid droplets: lipid droplets (totaln = 1404) were outlined using the 322 ( 32 S -/1) ratio value. The resulting ratio image was summed and processed with a gaussian blur 323 filter (sigma radius = 2 pixels). A threshold was applied to mask the image. ROIs were 324 identified by particle analysis, and verified by comparison with the respective SEM image. 325 Masked areas were overlaid to the 79 Br -/ 12 C 14 Nratio image of the same area of tissue. ROIs 326 with less than 5 pixels in size were excluded from the analysis. Mean normalised pixel intensity 327 in arbitrary units (AU) per ROI was plotted against condition using GraphPad Prism 8 software. 328 Two-tailed p-value was determined by an unpaired, non-parametric Mann-Whitney U test to 329 assess statistical significance. 330 331 Quantification of BDQ in bacteria inside LD: bacteria (totaln = 282) were manually outlined and 332 localisation defined to be either inside lipid droplets (inLD) or outside lipid droplets (outLD) 333 with the assistance of SEM images and the 31 Pand 32 Ssignal. Ratio values ( 79 Br -/ 12 C 14 N -) 334 for bacteria were divided by the area of their respective ROI to give mean normalised pixel 335 intensity in arbitrary units (AU) for each condition.