Development of Gram Stain Scoring System Based on Pro-Inflammatory Cytokines in the Sheep Model for Testing Toxicity of Vaginal Products

Abstract

Background: Development of safe, effective products to prevent the sexual transmission of HIV remains a priority. Prior to clinical testing, the products must undergo strict safety evaluations to avoid mucosal drug toxicity, inflammation, and vaginal microbiome (VMB) shifts. Based on the Food and Drug Administration (FDA) guidance, we designed a study to measure the inflammatory markers and VMB changes after intravaginal treatment with products that have been associated with toxicity, with the objective to develop a Gram stain slide scoring system, similar to Nugent scoring, correlated with the proinflammatory cytokines in sheep. Methods: Non-pregnant Dorset ewes (n = 34) were randomized to receive 5 ml intravaginal 4% nonoxynol-9 (N9) contraceptive gel, positive control (0.2% benzalkonium chloride), placebo control [hydroxethyl cellulose (HEC)], or no application daily for 10 days, with 11-day post-treatment follow-up. The vaginal swabs were collected for the cytokines, VMB, and Gram-stained slides. An enzyme-linked immunosorbent assay (ELISA) analysis of cytokines interleukin (IL)-1β, IL-8, CXCL10, and tumor necrosis factor-α (TNF-α) was used to determine inflammatory state of the sample. Vaginal microbiome community types (CT) were utilized to create five equivalent slide subsets for iterative development of a Gram-stained slide scoring system with comparisons with inflammatory state based on the cytokine levels. Results: Digital images of the Gram-stained slides were scored based on Gram staining and morphology of bacteria, presence of sheep epithelial cells, and immune cells. The scoring system was modified in an iterative fashion with weighting based on cytokine categorization of inflamed samples, with three of four cytokine values above the mean indicating that the sample was inflamed. The parameters in the final version of the scoring system included mature epithelial cells, Gram-negative rods, and Gram-positive diplococci indicating normal and immune cells indicating inflammation. The area under the receiver operator characteristic curve (ROC AUC) was 0.725 (ROC AUCs range between 0.5 and 1.0) with a greater area indicating higher diagnostic ability of a test with a binary outcome: inflamed or normal. Conclusion: The scoring system, derived from the advanced VMB and cytokine analyses, provides a validated, practical method for quantification of Gram-stained slides that can be performed in most laboratories, increasing the potential for standardization. The training plan can assist laboratories to determine the safety of intravaginal products in their sheep studies or the methodological approach can be applied to other animal models where such data are also needed.

Keywords: Gram stain scoring system; intravaginal drug delivery; ovine sheep vaginal model; proinflammatory cytokines; toxicity testing; vaginal microbiome.

Figure 1

Sheep inflammatory and epithelial cells. (A) Inflammatory cells can be seen in this Gram-stained slide as small cells with large or multi-lobulated nuclei. The epithelial cells were seen as (B) mature, which are the superficial cells with small nuclei and large cytoplasm, as (C) intermediate, which have less cytoplasm and slightly larger nuclei, and as (D), parabasal, which are near the basal or deepest cell layer and that have large nuclei and small amount of cytoplasm (micrographs are ×100 oil immersion).

Figure 2

Gram-positive bacteria seen in Gram-stained slides of sheep vaginal secretions. (A) Gram-positive cocci (left dashed box and magnified in insert on left) and Gram-positive diplococci (right dashed box and magnified in insert on right), (B) Gram-positive rods (left dashed box and magnified in insert on left) and pleiomorphic appearance of Gram-positive bacteria (right dashed box and magnified in insert on right). There was poor predictive ability for Gram-positive cocci, rods, and pleiomorphic bacteria, therefore they were not included in the final reduced scoring system (micrographs are ×100 oil immersion).

Figure 3

Gram-negative bacteria seen in Gram-stained slides of sheep vaginal secretions. (A) Gram–negative cocci (within circle) were very difficult to visualize due to pale staining and potential for confusion with debris. There was not good concordance between the graders, and this was not included in the final scoring system. (B) Gram–negative rods (within circle) were associated with a normal, non-inflammatory state (micrographs are ×100 oil immersion).

Figure 4

Iterative development of Gram stain slide scoring system. Vaginal microbiomes for each sample were determined by next generation sequencing (NGS) and quantitated PCR (qPCR), were categorized into community types (CT). Gram-stained slide images were split into five subsamples (A–E) with equivalent distribution of CTs. A scoring system (SS) was created based on counting the white blood cells (WBC), epithelial cells (EC), and Gram staining and morphotypes of bacteria, and applied to Subsample A. The counts were compared with cytokine-based inflamed categorization for weighting and modification of the SS. The modified SS was utilized for scoring Subsample B, and a similar process of comparison to inflammation, weighting, and modification was applied for each subsequent Subsample. After all the subsamples were analyzed with the modified SS used for Subsample E, and noted to have area under the curve (AUC) 0.69, the SS was simplified and reduced with removal of the conflicting parameters. The final reduced SS had AUC 0.725.

Figure 5

Subsample A receiver operating curve (ROC) area under the curve (AUC) analyses with v2.1 scoring algorithm showing acceptable sensitivity and specificity. After each subsample was scored and compared with inflammation as determined by cytokine analysis, an ROC AUC was calculated to determine the optimal scoring algorithm to be used for evaluation of the next subsample (_sens_, sensitivity; _spec_, specificity; _correct, proportion correctly categorized as “inflamed” or “normal,” at each cut-off). The red Y indicates the Youden, optimized, and cut-off.

Figure 6

The average ROC AUC values plotted across the study showing the potential for a matured epithelial cell and immune cell, Gram-positive diplococci, and Gram-negative rod reduced scoring system. Both the final expanded and reduced systems offer acceptable predictability, however, the reduced scoring system offers a slightly higher ROC AUC (±95% CI). The final reduced system is simplified, scoring the most easily recognized cells and morphotypes in the current set of Gram stains supporting its utility in future work (AUC is interpreted as follows: 0.5 AUC, no predictive ability; 0.7–<0.8, acceptable predictive ability; 0.8–<0.9, excellent predictive ability). The black circles indicate final expanded scoring system which included Gram-positive cocci, Gram-positive rods, Gram-positive diplococci, Gram-negative cocci, Gram-negative rods, inflammatory cells, and mature epithelial cells. The gray squares indicate the final reduced scoring system which included Gram-positive diplococci, Gram-negative rods, inflammatory cells, and mature epithelial cells. The Youden Index, which is the cut-off above which is considered inflammation, is shown in the table under the graph, with indicators of subsample and operator (e.g., A1 is subsample A graded by Operator 1).