Nipah Virus-Associated Neuropathology in African Green Monkeys During Acute Disease and Convalescence

Abstract

Background

Nipah virus is an emerging zoonotic virus that causes severe respiratory disease and meningoencephalitis. The pathophysiology of Nipah virus meningoencephalitis is poorly understood.

Methods

We have collected the brains of African green monkeys during multiple Nipah virus, Bangladesh studies, resulting in 14 brains with Nipah virus-associated lesions.

Results

The lesions seen in the brain of African green monkeys infected with Nipah virus, Bangladesh were very similar to those observed in humans with Nipah virus, Malaysia infection. We observed viral RNA and antigen within neurons and endothelial cells, within encephalitis foci and in uninflamed portions of the central nervous system (CNS). CD8⁺ T cells had a consistently high prevalence in CNS lesions. We developed a UNet model for quantifying and visualizing inflammation in the brain in a high-throughput and unbiased manner. While CD8⁺ T cells had a consistently high prevalence in CNS lesions, the model revealed that CD68⁺ cells were numerically the immune cell with the highest prevalence in the CNS of Nipah virus-infected animals.

Conclusions

Our study provides an in-depth analysis on Nipah virus infection in the brains of primates, and similarities between lesions in patients and the animals in our study validate this model.

Nipah virus (NiV) is a zoonotic virus that causes severe respiratory disease and meningoencephalitis [1]. A majority of NiV patients develop neurological disease acutely or weeks to even years after infection [2–5]. The pathophysiology of NiV meningoencephalitis is poorly understood, due to the low number of human autopsies and the lack of animal models focused on neurological disease. Common central nervous system (CNS) signs in human NiV cases include headaches, dizziness, and altered mental states [5–7]. Autopsies have only been performed during the NiV outbreak in Malaysia; the most common histologic lesion was vasculitis, occasionally with thrombosis, in multiple tissues including brain, lung, heart, and kidney. Syncytia formation was occasionally observed in endothelial cells, and in the CNS cytoplasmic and nuclear viral inclusions were observed [7, 8]. All autopsies were performed on patients who were placed on mechanical ventilation and other supportive care for at least 2 days [7, 8]. The impact this treatment had on CNS lesions is unknown.

Severe neurological disease is uncommon in the African green monkey (AGM) model, the preferred nonhuman primate model for NiV [9]. At our institute, we have collected the CNS of AGMs during multiple NiV studies, resulting in a unique collection of 14 brains that were shown to have NiV-associated lesions. Our study group contains animals euthanized with acute NiV disease, with animals having reached end point criteria with severe respiratory or neurological signs, and animals who survived lethal NiV challenge without showing signs of disease, making this pool of animals a valuable resource to study NiV neuropathogenesis. Here, we describe the spectrum and distribution of CNS lesions in AGMs during the acute and convalescent stage, and provide valuable insight into acute NiV meningoencephalitis.

METHODS

Ethics

This retrospective study used material from 3 studies of AGMs inoculated with NiV-B [10–12] and 1 study with mock-inoculated control AGMs [13]. Animal experiments were approved by the Institutional Animal Care and Use Committee of Rocky Mountain Laboratories, National Institutes of Health and carried out by certified staff in an Association for Assessment and Accreditation of Laboratory Animal Care international accredited facility, according to the institution's guidelines for animal use, following the guidelines and basic principles in the Guide for the Care and Use of Laboratory Animals, the Animal Welfare Act, United States Department of Agriculture and the United States Public Health Service Policy on Humane Care and Use of Laboratory Animals. Sample inactivation was performed according to standard operating procedures for removal of specimens from high containment approved by the Institutional Biosafety Committee.

Histology

Whole brains were fixed for a minimum of 30 days in 10% neutral-buffered formalin. Formalin-fixed brains were placed in a matrix mold and 8 evenly spaced coronal sections were collected. Hematoxylin and eosin (H&E) staining, in situ hybridization (ISH), and immunohistochemistry (IHC) were performed on formalin-fixed paraffin-embedded tissue sections. Detection of NiV viral RNA was performed using the RNAscope VS Universal assay (Advanced Cell Diagnostics) as previously described [14, 15]. Antibodies used for IHC are listed in Supplementary Table 1, and images from positive and negative control slides are available in Supplementary Figure 1. IHC was carried out on a Discovery ULTRA automated-staining instrument (Roche Tissue Diagnostics) with a Discovery ChromoMap DAB kit (Ventana Medical Systems).

UNet Model Development

We developed new deep learning models for positive cell segmentation on AGM brain tissue IHC scans. The models were based on a UNet [16] model architecture and the training set included 402 image patches of 1024 × 1024 in pixel size extracted from whole-slide images (WSI) with 40 × magnification. Image patches were selected to have balanced representation of positive and negative signals and imaging artifacts. Based on our pretrained UNet model [17], we created the training set annotation pixel masks through iterative transfer learning, using the pretrained model to predict a subset of training patches. At each iteration, results from the last iteration were cleaned up to remove false negatives and false positives, as determined by manual review by a pathologist (KG). The refined models were used to predict a set of unused training patches to produce more pixel masks. The training was done using a 3-fold cross validation schema. Each fold was trained up to 200 epochs with a learning rate of 1e-5, Adam optimizer, and sigmoid as the last activation layer. Image patches with a 256 × 256 pixel size were randomly extracted from the training images as the model training input images, with an up-sampling factor of 16 using nonoverlapping sliding window with stride size 0 as the base number to ensure good image feature coverage. Training image patches went through image augmentation steps including hue shifting, brightness and contrast adjustment, flipping, and rotation. The initial average validation score of 0.77 improved to 0.94 in the final average validation score. Then, a set of 10 WSIs were predicted using the new models to produce whole-slide pixel masks for visual examination by a pathologist.

To further validate the accuracy and robustness of our models, a second set of 1–3 regions of interest (ROIs) on 9 WSIs containing all markers used in the study were chosen by a pathologist (KG), who manually counted positive cells in the ROIs. Finally, our new UNet models were used to perform positive cell segmentation on the ROIs. For additional validation, QuPath 0.4.3 [18] was used to perform positive cell detection on the ROIs with default parameters except the “requested pixel size” parameter set to 0.25 µm ensuring the UNet models and QuPath ran at the same magnification. QuPath maximum nuclei area size was set to 400 µm². The same size limit was applied to the cell counting process after our UNet model prediction. Labels with area sizes larger than the threshold in connected component analysis were considered cell aggregates and the total aggregation area was divided by the threshold to estimate the cell counts. Finally, density maps were produced by counting positive cells in each area size of 0.1 × 0.1 mm². Our UNet models were run on NIH Biowulf. Positive cell counts from human expert manual counting, cell counting based on our UNet models, and QuPath positive cell detection were compared (Supplementary Table 2 and Supplementary Figure 2).

Statistical Analysis

Semiquantitative scores from the pathologist (KG) and UNet were compared using a pairwise Spearman correlation calculated using the pairwise_corr function in Pingouin package (version 0.5.4) in Python (version 3.9). Repeated measures correlation to account for multiple measurements for the same animals was calculated using the rm_corr function in Pingouin.

Data and Code Availability

The trained UNet models, WSI prediction, positive cell count, and density map creation python code are available in our GitHub repository (https://github.com/IDSS-NIAID/NiV_neuropath).

RESULTS

Nipah Virus-Associated Lesions Are Present in the Brains of Animals With and Without Overt Neurological Signs

We retrospectively analyzed 48 brains from AGMs inoculated with NiV-B for NiV-associated lesions; 2 brains from uninfected AGMs were included as controls. Of the 48 brains examined on H&E, pathologic lesions were detected in 14 animals (Table 1). Lesions were detected in animals with clear neurological signs, such as ataxia, hindlimb paresis, twitching in the limbs, head tilt, and, in 1 animal, anisocoria, and in animals where no overt neurological signs were observed (Table 1). Additionally, among the 34 brains without NiV-associated lesions, 8 were from animals presenting with mild to moderate neurological signs. This is likely the result of only examining 8 coronal sections per animal. The animals in which lesions were detected were all treated with remdesivir or vaccinated (Table 1), while no lesions were observed in untreated, infected control animals (n = 16). This is likely due to the rapid progression of lethal respiratory disease in untreated animals while treated animals survived longer allowing the progress of virus infection in the CNS, as was also observed in AGMs infected with Hendra virus and treated with ribavirin [19]. Lesions were observed in animals euthanized during acute disease, as well as in animals who survived lethal NiV challenge (Table 1). On histologic examination of H&E-stained sections, the most common finding was nonsuppurative meningitis, characterized by perivascular cuffing (Figure 1A). Rarely, blood vessels within foci of inflammation were undergoing fibrinoid necrosis (Figure 1B). While a common finding in human autopsy cases, fibrin thrombi were observed in a low number of necrotic vessels. There were occasional foci of encephalitis, characterized by infiltrating lymphocytes and macrophages, spongiosis, hemorrhage, and malacia (Figure 1C and 1D). A portion of inflammatory foci were composed predominantly of gitter cells, a phagocytic microglia subset characterized by abundant foamy cytoplasm, which often contains cellular debris, suggesting lesions were at least several days old, given the time it takes for infiltrating inflammatory cells to be cleared and resident microglia to migrate to the lesion (Figure 1E). A third type of lesion, which has been previously described in the literature as a “necrotic plaque” in both humans and AGMs [20, 21], was also frequently observed. These lesions are characterized by discrete focal loss of neuroparenchyma and normal resident cells (Figure 1F) and likely represent resolved foci of encephalitis and malacia, where the lost parenchyma and resident cells cannot be replaced once inflammation resolves. Because these lesions are not actively undergoing necrosis, as evidenced by the absence of hypereosinophilia, indistinct cell borders, and loss of cellular architecture, we feel that “necrotic plaques” does not accurately describe these lesions, and we will refer to them here as focal regions of parenchymal loss. Encephalitis, vasculitis, hemorrhage, and malacia were more common in animals who reached end point criteria compared to survivors. Although the foci of parenchymal loss were also seen in animals who were euthanized <18 days postinfection (dpi), they were more prevalent in animals who survived until end of study, supporting the hypothesis that these lesions are previously encephalitic foci that have since resolved. Perivascular cuffing was a common finding in both groups. Endothelial syncytia, while a hallmark of paramyxoviruses, were rare in animals who were euthanized <18 dpi, and none were observed in survivors. Similarly, intracytoplasmic and intranuclear viral inclusions are a typical finding in paramyxoviruses; however, we did not observe any that could be definitively identified.

Nipah Virus-Associated Lesions Have a Random Distribution Throughout the CNS

To determine whether NiV preferentially travels to and replicates in specific regions of the CNS, we compiled and mapped all observed CNS lesions (Figure 2). Lesions were observed scattered throughout both gray and white matter in all major parts of the brain, including the cerebrum, cerebellum, and brainstem. The distribution of the lesions appeared to be random, with no parts of the brain being grossly overrepresented. Glial nodules and encephalitic foci were more often found in the grey matter, consistent with previous reports in human autopsies [21], while perivascular cuffs and vascular lesions were evenly distributed. Lesions were more prevalent in animals who reached end point criteria than survivors (Figure 2, Supplementary Figure 3, and Supplementary Figure 4).

Nipah Virus Antigen and RNA Can be Found in Both Inflamed and Uninflamed Regions of the CNS

To gain better insight into the nature of the CNS lesions associated with NiV infection, we focused additional analyses on the CNS of 8 representative animals (Table 1), who were divided into 2 groups: those euthanized during acute NiV disease and survivors until study end with or without neurological signs. Using ISH and IHC, NiV RNA and N protein was detected in the CNS of all 8 animals. In the animals euthanized during acute NiV disease, NiV RNA and antigen were detected more often than in survivors. Overall, NiV RNA and antigen were observed multifocally throughout multiple sections in animals euthanized during acute disease, whereas viral RNA and antigen were rare in survivors. Multiple cell types were positive for both viral RNA and N protein antigen within the CNS, both within encephalitic foci and in uninflamed areas (Figure 3). Interestingly, no viral RNA or antigen was observed in the perivascular cuffs, despite this being the most common lesion observed on H&E. Viral RNA and antigen was observed in endothelial cells wherein there was no surrounding perivascular cuff and the endothelial cells were flat and nonreactive. Viral RNA and antigen were observed in neurons, both within the soma and processes. In encephalitic foci, viral antigen was observed within astrocytes on IHC (Supplementary Figure 5). Immunoreactivity was not as abundant within astrocytes as it was within neighboring neurons. Within the focal areas of parenchymal loss, there were often foci of positivity on both ISH and IHC. In 1 animal (AGM5), viral antigen was focally observed within the ependymal cells surrounding the third ventricle and within the underlying parenchyma (Supplementary Figure 6). Ependymal cells control flow and homeostasis of cerebrospinal fluid (CSF) [22]. This animal reached end point criteria at 9 dpi, and viral RNA was detected in the CSF (5.26 log₁₀ RNA copies/mL) and blood (4.93 log₁₀ RNA copies/mL). This was the only periventricular lesion in this animal (Supplementary Figure 3). Infection of the ependymal cells could have occurred via virus in the CSF entering the cell and extending to the underlying parenchyma or the infection could have extended up from the parenchyma to the ependyma. Regardless, viral RNA in the CSF provides evidence for a third method of viral dissemination throughout the CNS, in addition to hematogenous spread and accension via cranial nerves.

CD8⁺ Lymphocytes Are the Most Prevalent Cell Type in Inflammatory Lesions

We next assessed leukocyte populations in the CNS of the 8 representative animals using IHC (Supplementary Table 3). In perivascular cuffs and encephalitic foci, the dominant cell type was CD3⁺ T cells (Figure 4). The majority of these CD3⁺ cells were CD8⁺, rather than CD4⁺, T cells. The number of CD68 positive cells was quite variable within encephalitic foci. CD68⁺ cells made up a small percentage of cells in the perivascular cuffs. CD68⁺ cells may represent microglia or macrophages; unfortunately, attempts to further characterize these cell types using additional cell markers were unsuccessful. CD20⁺ B cells and CD4⁺ T cells were rare in both lesion types.

Development of a UNet Model for Generating Accurate Positive Cell Counts and Density Maps

To quantify the inflammatory response, a UNet model of image segmentation was developed. This UNet model generated positive pixel counts for leukocytes, which was in turn transformed into a positive cell count (Figure 5A). Importantly, IHC was also performed on the same representative sections from 2 uninfected control animals (AGM15, AGM16) because low numbers of leukocytes can be observed on IHC in healthy tissue sections within the meninges and the lumen of blood vessels. The UNet-generated cell counts were compared to the semiquantitative scores assigned by a veterinary pathologist (Figure 5B). For each antibody, a Spearman correlation was performed. There was a significant correlation between the UNet-generated cell counts and pathologist-assigned semiquantitative scores for CD8 (r = 0.8, P = .0058; Figure 5C). Interestingly, while CD8⁺ cells appeared to be the most prevalent on visual observation due to their consistent presence in inflammatory lesions, our quantitative UNet analysis showed that CD68⁺ cells were numerically the cell type with the highest prevalence, highlighting the utility of applying unbiased quantitation strategies to WSI. For the remaining antibodies, correlation was not significant, which may be attributed to both the low number of positive cells in the case of CD4 and CD20, and the relatively small group of WSI that this model was trained on. The overall correlation between the UNet-generated cell counts and the pathologist's scores for all antibodies were compared using repeated measure correlation and was found to be significant (r = 0.62, P = .00002; Figure 5C).

To better visualize the quantitative data generated by our UNet model, we used the positive cell identifications to generate a density map representing CD3⁺ positive cells in each area size of 0.1 × 0.1 mm² of the WSI. This density map clearly shows the focus of encephalitis in the brainstem and the meningitis observed by the veterinary pathologist on H&E of AGM6 (Figure 6). However, in contrast to a tissue slide observed through a microscope or cell counts in a graph, the density map provides a clear visual illustration of the entire tissue slide at once, simultaneously incorporating spatial as well as quantitative data.

DISCUSSION

AGMs are often presented as the gold standard model for NiV infection; however, overt neurological disease is not often observed in these animals as severe respiratory distress progresses more rapidly [23–25]. By combining animals from different studies, we can perform an in-depth analysis of NiV-B infection in the CNS of AGMs encompassing a broad spectrum of neurological disease signs. Unfortunately, autopsy studies on patients who succumbed to NiV-B infection are unavailable. We can thus not compare lesions in our AGMs to those in patients who died of NiV-B disease; detailed descriptions of CNS lesions in AGMs infected with NiV Malaysia are also lacking, preventing comparison of the 2 genotypes in AGMs. We found a high similarity between lesions observed in AGMs infected with NiV-B and the lesions in human cases of NiV-M [7, 8], suggesting that CNS lesions caused by both genotypes are likely very similar. This similarity enabled us to use the CNS from AGMs for an in-depth analysis of NiV infection of the CNS. In the absence of an animal model that consistently presents with overt neurological disease after NiV infection, retrospective, combined studies such as this are one of the few tools we have to increase our understanding of NiV neuropathogenesis. Minor differences observed between humans and AGMs may be related to differences in the time tissue was acquired: while AGMs were euthanized when they reached end point criteria, human patients received supportive care, likely prolonging the course of disease.

The NiV-associated histologic lesions from 14 AGMs were randomly distributed, indicating that NiV can spread widely throughout the CNS after entry. Dissemination appears to occur via interneuronal spread, as evidenced by the foci of neurons with NiV-positive cell bodies and neuronal processes, as well as via the CSF, as indicated by the presence of NiV in ependymal cells [20, 26, 27]. It is still unclear how NiV entered the CNS in these animals, whether via cranial nerves [14, 28], the hematogenous route [21, 29, 30], or both. We observed NiV antigen and RNA within endothelial cells, which may indicate hematogenous entry; however, because inflammation around these positive endothelial cells was rare, this may indicate that these cells do not get infected until late in disease. In that case, a dual entry of NiV via cranial nerves early after inoculation and via the blood once viremia occurs, as has been suggested for pigs inoculated with NiV-M [26], is likely.

NiV antigen and RNA was detected in inflamed and noninflamed regions, including within neurons and endothelial cells. The noninflamed regions could be early foci of virus replication that have not been reached yet by immune cells. However, because some of these areas were observed in animals euthanized 40 days after NiV inoculation, this appears less likely. Alternatively, this could be due to immunomodulation by NiV. Neurons are capable of producing type I interferons in response to viral infection [31]. In in vitro studies, several NiV proteins are capable of interfering with type I interferon expression by inhibiting the Toll-like receptor and RIG-I-like receptor pathways, including NiV-M [32], NiV-V [33–35], NiV-C [36], and NiV-W [37] proteins. Additionally, NiV-N [38], NiV-P [34], NiV-V [34], and NiV-M [32] can disrupt the JAK/STAT pathway and inhibit interferon-stimulated gene (ISG) expression, thereby dampening the host response. Findings from in vivo studies using recombinant NiV in hamsters showed that knocking out NiV-V or NiV-C results in a reduction in virulence; knocking out NiV-W had no effect on mortality [39]. In ferrets, knocking out NiV-C reduced respiratory disease, but had no effect on mortality [40], knocking out NiV-V resulted in a nonlethal virus [41], and knocking out NiV-W delayed onset of disease and increased neurological disease [41]. Although the effect of knocking out these proteins on histologic lesions in the CNS was not described, increased neurological signs and IHC positivity in the CNS was reported in animals where C and W proteins were both knocked out, suggesting that immune modulation by NiV affects NiV neuropathology.

Histopathology is often a rate-limiting step in animal studies, due to the time-intensive work of reading several slides per tissue per animal and limited availability of highly trained pathologists. Our UNet model can automatically accurately quantify positive IHC staining in tissue slides. This model could be deployed for any formalin-fixed paraffin-embedded tissue and multiple antigens with minimal modifications of the model and requires minimal equipment. The ability to perform IHC, a slide scanner, and the computational power to analyze whole slides is all that is required to generate cell numbers. In addition to cell counts, the ability to view the data in the form of a density map provides important spatial information. The density map is an intuitive, easy way to interpret the images, even for researchers without pathology training. This can be used as a screening tool in conjunction with a pathologist, wherein a set of WSI could be analyzed, density maps generated, and slides identified with lesions of interest that may require manual examination by a pathologist.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online (http://jid.oxfordjournals.org/). Supplementary materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supplementary data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author.

Notes

Acknowledgments. The authors thank Dan Long, Tina Thomas, Greg Saturday, and Carl Shaia for help with histology; Jacqueline Leung for help with data sharing; and Ryan Kissinger and Rose Perry-Gottschalk for help with figure preparation.

Author contributions. K. G. and E. d. W. conceived experiments. K. G., Y. L., R. R., and J. P. S. carried out experiments. K. G., Y. L., M. F., and E. d. W. analyzed data. K. G. and E. d. W. wrote the first draft of the manuscript. All authors reviewed and edited the manuscript.

Financial support. This work was supported by the Intramural Research Program, National Institute of Allergy and Infectious Diseases, National Institutes of Health.

References

Author notes