2.1.

Mice and Tissue Sample Preparation

Female C57BL/6 mice were purchased from the Centre for Animal Resources, Singapore and housed in specific pathogen-free BSL2 facilities at the National University of Singapore (NUS). Mice were anesthetized with ketamine ($100\mathrm{mg}/\mathrm{kg}$ body weight) and infected with a sub-lethal dose (100 PFU/mouse) of the influenza virus A/Puerto Rico/8/34 (H1N1) by intratracheal instillation. Five mice were sacrificed at each time point: before infection (control) and 1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, and 28 days post infection (dpi). Lung tissues were collected. The left lobes were fixed in 10% neutral buffered formalin solution (Sigma-Aldrich, HT501128-4L) overnight and processed with Tissue Processor (Leica, TP1020, Shanghai, China) and embedded in paraffin blocks. For each lobe, 15 transverse sections were cut with microtome (Leica, RM2165, Nussloch, Germany) from the middle part of the lobe and with 50 μm in between sections. Sections were mounted on polylysine-coated slides (Thermal Fisher Scientific, J2800AMNZ). Ten sections were used for H&E and five sections for immunofluorescent staining. The right lobes were used for lavage, measuring virus titer and other analysis. All animal protocols were approved by the Institutional Animal Care and Use Committee at NUS and Massachusetts Institute of Technology.

2.2.

RNA Extraction and Real-Time Fluorescence Quantitative PCR

Mice were sacrificed at indicated time points post infection, and total RNAs were extracted from lung tissues with TRIZOL reagent (Life Technologies, 15596–026) according to the manufacturer’s protocol. Two micrograms of total RNA from each sample was reverse-transcribed using oligo(dT) primers (Promega, C1101) and Superscript II reverse transcriptase (Life Technologies, 18064–014). PCR was performed with SsoFast EvaGreen Supermix (Bio-Rad, 172–5210) using the CFX96 real-time PCR detection system (Bio-Rad Laboratories, C1000 Thermal Cycler, Singapore). Primer sequences for the L32 housekeeping gene and influenza virus N gene were chosen as previously described.^28,29

2.3.

Histological Staining

Paraffin sections were de-waxed in Histo-Clear solutions twice (National Diagnostics, HS-200) and rehydrated first in absolute ethanol three times, and then once each in 90% ethanol, 70% ethanol, and 50% ethanol. H&E staining was processed according to standard protocol. For immunofluorescent staining, antigen retrieval was performed by incubating the lung section with proteinase K solution (Sigma-Aldrich, $20\mathrm{mg}/\mathrm{ml}$, in 50 mM Tris-Cl, 1 mM EDTA, pH 8.0) at 37°C for 30 min. Sections were then blocked in blocking buffer (3% BSA, 0.2% Triton X-100 in PBS) for 1 hr. Polyclonal rabbit anti-CCSP (also known as Scgb1a1) antibody (USbiology, C5828) was used in $1:200$ dilution, and goat anti-pro-SPC (Santa Cruz sc-7706) antibody was used in $1:50$ dilution. Incubation was performed at 4°C overnight in blocking buffer. Secondary Alexa Fluor 488-labeled donkey anti-rabbit antibody (Invitrogen, A21206) or Alexa Fluor 546-labeled donkey anti-goat antibody (Invitrogen, A11056) were used in $1:200$ dilution. Incubation was performed at 4°C for 1 h. Cover slips were mounted on stained section with antifade reagent containing DAPI (Invitrogen, S36939).

2.4.

Image Acquisition

A high-resolution MIRAX MIDI system (Carl Zeiss, Jena, Germany) equipped with bright field and fluorescence illumination was used to scan the stained lung sections. Images were captured with Axiocam MR(m) (Carl Zeiss, Thronwood, New York), and converted to TIFF format with Miraxviewer software (Carl Zeiss). The original resolution of all images is $1.86\times 1.86{\mu \mathrm{m}}^2$ per pixel. The images used for quantification were down-sampled eight times from the original images to reduce the computation time and the use of memory space.

2.5.

Histology Analysis

Digital images of H&E stained lung samples from different time points post infection were evaluated blindly by three independent researchers. A semi-quantitative grading system was applied as previously reported,⁹ in which subjective scores from 0 to 3 were given to reflect absent, mild, moderate, and extensive cell infiltration, respectively. Scores from the three researchers were averaged to obtain the average score of cell infiltration at each time point post infection.

2.6.

Feature Extraction and Quantification

Automated algorithms were developed to compute four indices for quantitative evaluation of lung damage and repair based on image analysis. All image processing and computation algorithms were implemented with Matlab with the image processing toolbox (The Math Works Inc., Natick, Massachusetts). The Matlab codes are available on request.

2.6.1.

Infiltration index

The infiltration index was defined as the percentage of infiltrated areas to total alveolar area in H&E stained lung section images. The infiltration index was calculated by infiltration algorithm as follows. The original H&E stained RGB image was first converted to greyscale. The image pyramid of greyscale image was then created by subsampling the image by a factor of 1, 2, 4, and 8, respectively, along each spatial dimension.³⁰ A threshold was manually adjusted and applied to identify low intensity pixels in each level of the image pyramid. A pixel was identified within the infiltrated area if it was selected as low intensity pixels in all levels of the image pyramid. The mask of the infiltrated area was generated by dilating all pixels identified within the infiltrated area to connect small regions. The dilation was performed with a “disk” operator of 8 pixels in size. The total alveolar area was segmented from the original image by excluding large empty blood vessels and bronchioles with a threshold of 100 pixels in size. The percentage of infiltrated area to total alveolar area was then computed as the infiltration index and used as an index of general lung damage and repair at tissue level (Fig. 1).

Fig. 1

Flowchart of the feature extraction and quantification algorithm for infiltration index.

2.6.2.

Clara cell coverage index

The Clara cell coverage index reflected the coverage density of Clara cells on the interior wall of bronchioles. Bronchioles were automatically recognized from immunofluorescent images as empty spaces with no nuclei inside and surrounded by Scgb1a1-expressing Clara cells [Fig. 2(a)]. Nuclei were segmented from the original image by manual thresholding DAPI fluorescence intensities [Fig. 2(b)]. Centers of each nucleus were located and denoted as nodes. A delaunay triangulation diagram was generated by introducing all nodes to connect neighboring nodes while none was inside the circumcircle of any triangle formed³¹ [Fig. 2(c)]. Nodes at the boundary of bronchioles (bronchiolar epithelium) tended to have longer average edge length than others, and nonboundary nodes with shorter average edge length were neglected accordingly. Boundary nodes belonging to the same bronchiole were then clustered together based on their connectivity in the original triangulation. Centers of each boundary node cluster were located and region growing³² was performed from the centers to determine the exact boundaries of potential bronchioles [Fig. 2(d)]. To further recognize bronchioles from all candidates, Clara cells were identified from the original image by manual thresholding Scgb1a1 fluorescence intensities. The Clara cell coverage ratios around the boundary area were calculated for each candidate, and a minimum threshold (1%) was applied to exclude false ones with completely no Clara cells surrounded [Fig. 2(e)]. The remaining true bronchioles were then clustered into the high-Clara-cell-coverage ones (HC) and low-Clara-cell-coverage ones (LC) using the Otsu method³³ based on the Clara cells coverage ratio calculated. The Clara cell coverage index was defined as the ratio of HC frequency to LC frequency and used for the evaluation of the lung bronchiolar damage and repair.

Fig. 2

Illustration of algorithms for quantification of immunofluorescent images. (a) Snapshot of lung tissue area containing three bronchioles to be automatically identified. Nuclei were stained in blue (DAPI), Clara cells were stained in green (Scgb1a1), and AT2s were stained in red (pro-SPC). (b) Segmentation of nuclei in the snapshot spot. (c) Delaunay triangulation diagram generated with the centers of each nuclei to define boundary nodes around bronchioles. (d) Candidates of potential bronchioles identified after performing region growing from centers of clusters of boundary nodes. (e) Bronchioles identified after examining the presence of Clara cells along the boundaries. (f) Segmentation of AT2s in the snapshot spot. (g) Global view of the scanned image of one entire lobe of an infected mouse lung, consisting of a healthy area (HA) with normal alveolar structure and a damaged area (DA) with condensed infiltrating cells. (h) Damaged area from the tissue section segmented with $k$-means clustering algorithm. (i) Identified bronchioles containing $\mathrm{Scgb}1\mathrm{a}{1}^{+}$ (solid arrowheads) and $\mathrm{Scgb}1\mathrm{a}{1}^{-}$ (open arrowheads) SBECs.

2.7.

Statistical Analysis

Pearson correlation coefficients and $p$-values have been calculated to measure the consistency between: (1) cell infiltration kinetics obtained by automated algorithm and manual grading; (2) cell infiltration kinetics and Clara cell recovery kinetics from 5 to 14 dpi, and (3) Clara cell recovery kinetics and SBEC induction kinetics from 5 to 14 dpi. All statistics were computed with Matlab (The Math Works Inc., Natick, Massachusetts). The Matlab codes are available on request.

3.1.

Quantification of Cell Infiltration in the Infected Lung

The influenza virus infection induces immune responses in the infected mice, leading to immune cell infiltration into the lung. We first quantify the cell infiltration as a frame of reference for the damage and repair process. The infiltrating cells appeared as densely packed nuclei in H&E staining [Fig. 3(a)]. To quantify the cell infiltration in the infected lung, we developed an infiltration algorithm by segmenting dark infiltrated areas from the healthy areas in images of H&E stained lung sections, and computing the infiltration index by averaging the percentage of infiltrated area to total alveolar area in each lung sections (Fig. 1). By calculating the infiltration index over the 28 days following infection, we obtained a dynamic picture of cell infiltration [Fig. 3(b)]. The infiltration was clearly detected 5 to 7 days post infection (dpi), corresponding to the length of time required to induce a T cell response.³⁶ The level of infiltration increased continuously and peaked around 14 dpi. By 21 and 28 dpi, most of the infiltrating cells were gone, corresponding to recovery from infection. The levels of cell infiltration as calculated by the automated algorithm matched well with those obtained from a semi-quantitative grading system,⁹ in terms of overall kinetics (Pearson correlation coefficient equals to 0.964, $p=1\times {10}^{-7}$) and peak days [Fig. 3(c)].

Fig. 3

Quantification of lung damage and repair based on cell infiltration. (a) Representative H&E staining of the lung sections at the indicated time points post-influenza infection. (b) Infiltration index at the indicated time points post infection was calculated using the infiltration algorithm. (c) Infiltration score estimated at the indicated time points post infection using the classic semi-quantitative grading system. Data were from ten sections per mouse and five mice per time points. All error bars represent standard errors (SE).

Several imaging-based quantitative analyses of H&E stained lung sections have previously been reported. In the hyperoxia-induced lung injury where the network of alveolar sacs was destroyed, the quantification was based on the size of the alveolar sac.^15,16 In bleomycin-induced lung injury the quantification was based on the tissue thickness because of lung fibrosis.¹⁷ Following the influenza virus infection, cell infiltration into the lung is prominent; therefore we used the levels of cell infiltration as a measure of the lung damage and repair. Compared to the semi-quantitative grading,⁹ our automated algorithm has similar dynamic range and accuracy.

3.2.

Quantification of Clara Cells, AT2s, and SBECs

In healthy lung, Scgb1a1-expressing Clara cells cover the bronchiolar epithelium whereas pro-SPC-expressing AT2s are dispersed in the alveolar epithelium [Fig. 4(a) and 4(b)]. Influenza viruses infect Clara cells and AT2s directly³⁷ (Zheng et al. submitted), leading to gradual loss of both cell types. Loss of Clara cells was observed as early as 1 dpi (data not shown) and reached the peak level 5 dpi when Clara cells sloughed off from bronchiolar epithelium [Fig. 4(c) and 4(d), star]. From 7 dpi onwards, more Scgb1a1-expressing cells were detected in the bronchioles in the damaged areas [Fig. 4(e) and 4(f)], suggesting recovery of Clara cells. Loss of AT2s was observed as early as 3 dpi (data not shown) and large areas were depleted of AT2s by 5 and 7 dpi [Fig. 4(c) to 4(f)]. Pro-SPC-expressing cells started to become detectable in the damaged parenchyma 9 to 21 dpi, suggesting the regeneration of AT2s.

Fig. 4

Visualization of Clara cells, AT2s, and SBECs following the influenza virus infection. Representative immunofluerescent staining of lung sections for Scgb1a1 (green) and pro-SPC (red) at the indicated time points following influenza infection. Scale bars in the low resolution images (left panel) are 500 μm; the boxed areas were shown at high resolution (right panel) and the scale bars are 100 μm. The arrows point to loss of Scgb1a1-expressing cells; the solid arrowheads indicate regions with loss of pro-SPC-expressing cells; the open arrowheads point to SBECs; and the star indicates a bronchiole with sloughed off Clara cells. DAPI stain was not shown.

We have developed an automated algorithm to quantify the loss and recovery of Clara cells using immunofluorescent images of lung sections. Bronchioles were first identified in the images as closed empty structures with some Scgb1a1 signals (at least 1%) at boundaries (bronchiolar epithelium), and the coverage of Clara cell was then computed by calculating the ratio of bronchiolar epithelium length covered by ${\mathrm{Scgb}1\mathrm{a}1}^{+}$ cells to total bronchiolar epithelium length for each bronchiole [Fig. 2(a) to 2(e)]. The bronchioles were divided into the high-Clara-cell-coverage (HC) and low-Clara-cell-coverage (LC) bronchioles; and the ratio of HC to LC bronchioles was defined as the Clara cell coverage index. Calculation of the Clara cell coverage index over the 21 dpi revealed that Clara cells were lost from bronchioles immediately after infection, reached the peak loss 5 dpi, and then gradually recovered to normal level by 14 dpi [Fig. 5(a)].

Fig. 5

Dynamic changes of Clara cells, AT2s, and SBECs following the influenza virus infection as calculated by the computational algorithms. (a) Clara cells coverage index over time following infection. (b) AT2 density (number of AT2s $\mathrm{per}{\mathrm{mm}}^2$ tissue) over time following the infection. (c) Frequencies of $\mathrm{Scgb}1\mathrm{a}{1}^{+}$ and $\mathrm{Scgb}1\mathrm{a}{1}^{-}$ SBECs over time following the infection. All error bars represent SE.

Similarly, we quantified the loss and recovery of AT2s using immunofluorescent images of lung sections. Healthy areas in the lung without apparent cell infiltration were first distinguished from damaged areas by $k$-means clustering-based segmentation.³⁴ AT2s were then identified as pro-SPC-expressing cells in the healthy regions of the lung. The relative density of AT2s was calculated as the number of AT2s per unit tissue section area. Quantification of AT2 density over 21 days post infection showed that AT2s were gradually lost after infection and the most severe loss was around 7 dpi [Fig. 5(b)]. By 21 dpi, most of AT2s were regenerated in the damaged parenchyma.

SBECs were first detected around 7 dpi and became abundant by 14 dpi [Fig. 4(g) and 4(h)]. Some SBECs were ${\mathrm{Scgb}1\mathrm{a}1}^{+}$ whereas others were ${\mathrm{Scgb}1\mathrm{a}1}^{-}$ (Fig. 6). We have quantified the frequencies of ${\mathrm{Scgb}1\mathrm{a}1}^{+}$ and ${\mathrm{Scgb}1\mathrm{a}1}^{-}$ SBECs based on immunofluorescent images. Damaged areas in the lung with severe cell infiltration were first identified using $k$-means clustering-based segmentation. Bronchioles in the damaged areas were then identified using the same way in the calculation of the Clara cell coverage index. SBEC-containing bronchioles were identified as bronchioles containing pro-SPC signals in the damaged areas. Frequencies of ${\mathrm{Scgb}1\mathrm{a}1}^{+}$ and ${\mathrm{Scgb}1\mathrm{a}1}^{-}$ SBEC-containing bronchioles were quantified by examining the colocalization of Scgb1a1 and pro-SPC signals in the identified bronchioles in the damaged areas. As shown in Fig. 5(c), the ${\mathrm{Scgb}1\mathrm{a}1}^{+}$ SBECs first appeared around 7 dpi, peaked around 14 dpi, and decreased by 21 dpi. ${\mathrm{Scgb}1\mathrm{a}1}^{-}$ SBECs were not observed until 9 dpi and then followed the similar dynamic profile as ${\mathrm{Scgb}1\mathrm{a}1}^{+}$ SBECs, suggesting that ${\mathrm{Scgb}1\mathrm{a}1}^{-}$ SBECs come from ${\mathrm{Scgb}1\mathrm{a}1}^{+}$ SBECs.

Fig. 6

Representative immunofluerescent staining of lung sections for Scgb1a1 (green) and pro-SPC (red) 14 dpi. The boxed areas in (a) are shown in higher magnification in (b) and (c), showing $\mathrm{Scgb}1\mathrm{a}{1}^{+}$ (solid arrowheads) and $\mathrm{Scgb}1\mathrm{a}{1}^{-}$ (open arrowheads) SBECs. Scale bars are 500 μm in the low magnification image and 100 μm in the high magnification images.

3.3.

Insights into Lung Tissue Damage and Repair

We compared the kinetics of virus titers, cell infiltration, Clara cell and AT2 loss and recovery, and induction of SBECs in the lung during the 21 days following the influenza infection. For easy comparison, the values were normalized and plotted over the same time course (Fig. 7). First, cell infiltration into the lung continued even when most of viruses were cleared. Second, the continued cell infiltration preceded the Clara cell recovery. Third, the kinetics of SBEC induction starting at 7 dpi lagged behind but paralleled Clara cell recovery starting at 5 dpi (Pearson correlation coefficient equals to 0.9587, and $p=0.0487$ between the kinetics of Clara cell recovery and ${\mathrm{Scgb}1\mathrm{a}1}^{+}$ SBEC induction from 5 to 14 dpi. The Pearson correlation coefficient equals to 0.9763, and $p=0.0237$ between the kinetics of Clara cell recovery and ${\mathrm{Scgb}1\mathrm{a}1}^{-}$ SBEC induction from 5 to 14 dpi). Finally, in contrast to Clara cells and SBECs that peaked at 14 dpi, recovery of AT2s did not reach the peak level even 21 dpi. These results are consistent with the notion that Clara cells give rise to ${\mathrm{Scgb}1\mathrm{a}1}^{+}$ SBECs and then ${\mathrm{Scgb}1\mathrm{a}1}^{-}$ SBECs, which then differentiate into AT2s during the repair of damaged alveolar epithelium (Zheng et al. submitted). The continued cell infiltration into the lung when viruses were already cleared suggests that infiltrating cells likely contribute to tissue damage. The parallel kinetics between cell infiltration and recovery of Clara cells also raises the intriguing possibility that infiltrating immune cells may be involved in lung tissue repair. Thus, our systematic and quantitative analyses shed new light on lung tissue damage and repair following influenza virus infection.

Fig. 7

Comparison of (a) virus titer, (b) cell infiltration, (c) loss and recovery of Clara cells, (d) induction of $\mathrm{Scgb}1\mathrm{a}{1}^{+}$ and $\mathrm{Scgb}1\mathrm{a}{1}^{-}$ SBECs, and (e) loss and recovery of AT2s following the influenza virus infection. The values were normalized to the 0 to 1 range.

In this study, we report new algorithms for quantifying cell infiltration, Clara cells, AT2s, and SBECs during the influenza virus-induced lung damage and repair. The methods are based on images acquired with an automatic slide scanner, which can scan a large area, such as the entire transverse lung section, with high speed and adequate resolution. As a result, we could obtain a comprehensive picture over the entire tissue section and specific spatially related features. For example, two stages of SBECs could be accurately identified and quantified in bronchioles in the damaged lung areas. To obtain the same information with microscopy modalities would be impractical because large numbers of small areas of the lung have to be analyzed and then combined. Thus, our quantification algorithms take advantage of the images with high content from a slide scanner, and extract specific features over large areas of tissue for quantification. Furthermore, our analyses extract features, such as nucleus density and spatial distribution of molecular markers; the same features could be used for studying cell dynamics in lung injury induced by other agents with no or minor modifications of the existing methods. Our spatiotemporal approach of image analysis could be potentially applied in study of cellular mechanisms of other diseases.