Flower lose, a cell fitness marker, predicts COVID-19 prognosis

Abstract

Risk stratification of COVID-19 patients is essential for pandemic management. Changes in the cell fitness marker, hFwe-Lose, can precede the host immune response to infection, potentially making such a biomarker an earlier triage tool. Here, we evaluate whether hFwe-Lose gene expression can outperform conventional methods in predicting outcomes (e.g., death and hospitalization) in COVID-19 patients. We performed a post-mortem examination of infected lung tissue in deceased COVID-19 patients to determine hFwe-Lose's biological role in acute lung injury. We then performed an observational study (n = 283) to evaluate whether hFwe-Lose expression (in nasopharyngeal samples) could accurately predict hospitalization or death in COVID-19 patients. In COVID-19 patients with acute lung injury, hFwe-Lose is highly expressed in the lower respiratory tract and is co-localized to areas of cell death. In patients presenting in the early phase of COVID-19 illness, hFwe-Lose expression accurately predicts subsequent hospitalization or death with positive predictive values of 87.8-100% and a negative predictive value of 64.1-93.2%. hFwe-Lose outperforms conventional inflammatory biomarkers and patient age and comorbidities, with an area under the receiver operating characteristic curve (AUROC) 0.93-0.97 in predicting hospitalization/death. Specifically, this is significantly higher than the prognostic value of combining biomarkers (serum ferritin, D-dimer, C-reactive protein, and neutrophil-lymphocyte ratio), patient age and comorbidities (AUROC of 0.67-0.92). The cell fitness marker, hFwe-Lose, accurately predicts outcomes in COVID-19 patients. This finding demonstrates how tissue fitness pathways dictate the response to infection and disease and their utility in managing the current COVID-19 pandemic.

Keywords: COVID-19; biomarker; cell fitness; flower; prognosis.

Figure 1. hFwe‐Lose biomarker associated with COVID‐19 mortality and host comorbidities

1. A schematic of cell competition process. Our bodies have a natural surveillance system that optimizes tissue fitness. The process of cell competition drives healthy tissues to force suboptimal, yet viable, loser cells to undergo cell death. Various stressors and insults cause cells to alter their properties and expression of fitness biomarkers. Cellular fitness comparisons lead to the elimination of loser cells that express hFwe‐Lose, a biomarker of reduced fitness. This mechanism is responsible for actively restoring tissue homeostasis and has important implications in response to infections and the development of malignancies. ROS: reactive oxygen species.

2. hFwe‐Lose mRNA expression is more abundant in elderly people. hFwe‐Lose mRNA expression was analyzed by RT–qPCR in 86 lung tissue biopsies taken from non‐COVID patients with age between 20 and 82 years. Older patients show a significant upregulation of hFwe‐Lose expression. A log‐linear regression model demonstrates a positive correlation between age and hFwe‐Lose expression (R ² = 0.13; slope confidence interval of 95% (CI) = [2.0–12.2]; P‐value of the linear regression model < 6 × 10⁻⁴).

3. hFwe‐Lose expression is elevated in lung tissue biopsies from patients with comorbidities. Box plot illustrates an increased expression of hFwe‐Lose in lungs of patients with hypertension (HT; n = 129), obesity (n = 45), chronic obstructive pulmonary disease (COPD; n = 51), diabetes (n = 48), cardiovascular disease (CVD; n = 63) versus disease‐free control lungs (n = 42). Patient`s age is depicted in color. Two‐sided Student’s t‐test was performed for each comorbidity (compared to disease‐free patients), and P‐values are presented on the plot. The central band shows the median, the box indicates the interquartile range, and the whiskers extend to the most extreme points within the 1.5‐fold distance of the interquartile range above and below the box.

4. hFwe‐Lose expression is upregulated in lung tissue of COVID‐19 patients. Box plot illustrates an increased expression of hFwe‐Lose in lung tissue of patients diagnosed with COVID‐19 (n = 11), individuals affected with host comorbidities (n = 216) versus disease‐free control lungs (n = 42). Patient’s age is depicted in color. Two‐sided Student’s t‐test was performed (compared to disease‐free patients), and P‐values are presented on the plot. The central band shows the median, the box indicates the interquartile range, and the whiskers extend to the most extreme points within the 1.5‐fold distance of the interquartile range above and below the box.

5. SARS‐CoV‐2 infection manifests histological changes in patients' lungs as demonstrated by H&E staining. 1) For comparison, the normal lung of an elderly individual is shown, containing air‐filled empty looking alveolar spaces and thin alveolar septae/gas‐exchange membranes. Note that there are almost no inflammatory cells and that capillaries are only merely visible since not congested. 2) Diffuse (proliferative) alveolar damage showing evidence of cell death. An alveolar space filled with desquamated pneumocytes and macrophages, lymphocytes, and focal erythrocyte extravasation as well as one multinucleated pneumocyte type II. The still recognizable epithelial lining is detached. Cells and surfactant are lost creating perforations in the alveolar wall, allowing migration of blood cells and fluid to enter inside the alveolar space. The adjacent interstitial space is significantly widened, showing an increase in mononuclear inflammatory cells and extravasation of erythrocytes. On the top and the lower right, dilated and congested capillaries can be seen (hematoxylin and eosin (H&E) stain, 200×). a—alveolar structure destroyed by lymphocytes, desquamated epithelium, and extravasated blood; the alveoli should be empty, but here it is filled with a combination of degenerated cells and fibrin; b—next alveoli; c—expanded pulmonary interstitium. The interstitium should be as thin as 10 μm, but here it is 100 μm; d—extravasation of erythrocytes; e—lymphocytes; f—congested capillaries and arterioles; g—multinucleated alveolocyte/pneumocyte type II; h—detached epithelium; i—a histiocyte with a kidney‐shaped nucleus at the bottom of the circle has ingested erythrocytes (left‐handed), representing the first step of cellular elimination. Round inlet with encircled karyopyknotic, karyorrhectic, and “ghost‐cell” figures indicative of apoptosis. 3) The presence of type II pneumocyte syncytial giant cells in a collapsing alveolar space with detached epithelial lining. The adjacent interstitial space shows analogous changes to the previous example. At the bottom, there is a prominently dilated and congested capillary (H&E, 200×). a—alveoli with detached epithelium; b—next alveoli; c‐ Extravasation of erythrocytes; d—expanded interstitium; e—lymphocytes; f—multinucleated pneumocyte type II. 4) Diffuse alveolar damage showing massive extravasation of fibrin (homogeneous eosinophilic material in the center of the alveolar space). The lining pneumocytes are almost all apoptotic/necrotic. The fibrin exudate is intermingled with mononuclear inflammatory cells and cellular debris (H&E, 200×). a—alveola; b—congested capillaries; c—fibrin; d—remnants of degenerated/dying epithelium; e—expanded interstitium; f—cellular debris consisting of macrophages, detached epithelial cells and lymphocytes; g—fully degenerated/lacking epithelial coverage within the alveoli. Round inlet with encircled karyopyknotic, karyorrhectic, and “ghost‐cell” figures indicative of apoptosis. 5) Immunohistochemical stain (IHC) for fibrin showing microthrombi caused by dysfunction of endothelial cells in capillaries of the alveolar membranes/lung interstitium leading to obstruction of the microcirculation (IHC for fibrin, 200×). a—all small alveolar septal capillaries are filled out with worm‐like fibrin thrombi hampering / obstructing circulation; b—normal interstitium.

6. Healthy lungs show very low cleaved caspase‐3‐positive cells. IHC staining for cleaved caspase‐3 (20×).

7. Expression of cleaved (active) caspase‐3 yielding apoptotic and pre‐apoptotic cells: IHC for cleaved (active) caspase‐3 showing a brown nuclear staining signal in respective cells. 31.5% cells were found to be caspase positive, 11.7% with high caspase‐3 positivity and another 19.8% with low. These cells are mainly located in the interstitium, but also in some alveolar epithelial and endothelial compartments. In the lower right, the edge of an alveolar space containing several (pre‐)apoptotic pneumocytes is seen, while in the upper left, a completely denuded/deepithelialized alveolus with two apoptotic remnants is observable (IHC for cleaved caspase‐3, 20×). a—apoptotic large mononuclear cells; b—apoptotic alveolocyte / pneumocyte; i—amplification of apoptotic alveolocyte / pneumocyte. As shown in the round inlets of 1E2 and 1E4, homogeneous dark nuclear condensation up to a size of 2 μm (karyopyknosis), homogeneous pinkish‐grayish nuclear condensation, large nuclear inclusion with marginalization of the condensed chromatin, coarse nuclear angulation, nose‐like/polar body‐like nuclear protrusions, the latter two being karyorrhectic debris, were all morphologically considered evidence of apoptosis. This has been correlated with and was reflected by the results of the cleaved caspase‐3 staining on step sections. All cells were counted on the 20× fields for n = 42 disease‐free and n = 3 COVID‐19 autopsy patients. The expression of hFwe‐Lose in COVID‐19 patients is significantly higher than in disease‐free individuals, regardless of cleaved caspase‐3 staining (Cas POS: P < 0.004; Cas NEG: P < 0.006). In COVID‐19 patients, we observed a significantly higher expression of hFwe‐Lose in sections with positive cleaved caspase‐3 staining (P < 0.004). Two‐sided Student’s t‐test was performed (compared to disease‐free samples and NEG samples, respectively), and P‐values are presented on the plot. The central band shows the median, the box indicates the interquartile range, and the whiskers extend to the most extreme points within the 1.5‐fold distance of the interquartile range above and below the box.

Figure 2. hFwe‐Lose biomarker, measured in nasopharyngeal swab samples, associates with patients’ COVID‐19 disease outcome

1. hFwe‐Lose biomarker expression is more abundant in nasopharyngeal swab probes from older adults. hFwe‐Lose expression was analyzed by RT–qPCR in 283 nasopharyngeal swab samples taken from patients with age between 1 and 96 years, taken at the very beginning of the disease (the earliest contact with physician, before the disease progression). The vertical axis represents relative hFwe‐Lose expression normalized to the mean of non‐hospitalized patients. Colors depict the outcome groups: non‐hospitalized (gray, n = 85), hospitalized (blue, n = 177), and deceased (red, n = 21). The shape of data points reflects the cohorts: circles for the training cohort (n = 203) and triangles for the validation cohort (n = 80). The lines show the fitted curves of an asymptotic model with the same asymptotic value but different rate constants per group (see Materials and Methods). Due to the comparatively low number of deceased patients in the dataset (n = 21), the curve for this group reflects the asymptotic value. Hospitalized and deceased patients show a positive correlation of hFwe‐Lose expression and age with a larger rate constant for the hospitalized patients (R ² = 0.65). The P‐value (< 0.001) indicates that the blue curve (for hospitalized patients) grows faster with age, compared to the gray curve (for non‐hospitalized patients).

2. hFwe‐Lose expression is elevated in nasopharyngeal swab probes from patients with comorbidities. Box plots illustrate an increased relative expression of hFwe‐Lose in nasopharyngeal swabs of patients with diabetes (n = 129), COPD (n = 20), obesity (BMI > 30; n = 152), cardiomyopathy (CM; n = 19), heart failure (HF; n = 35), hypertension (HT; n = 121), chronic kidney disease (CKD; n = 60) versus disease‐free patients (n = 96). Two‐sided Student’s t‐tests were performed (compared to disease‐free patients), and P‐values are presented on the plot. The vertical axis represents relative hFwe‐Lose expression normalized to the mean of non‐hospitalized patients. The color refers to the COVID‐19 disease outcome: gray for not hospitalized, blue for hospitalized and red for deceased patients. The shape of data points reflects the cohorts: circles for the training cohort (n = 203) and triangles for the validation cohort (n = 80). The central band shows the median, the box indicates the interquartile range, and the whiskers extend to the most extreme points within the 1.5‐fold distance of the interquartile range above and below the box.

3. An age‐ and sex‐adjusted statistical model suggests hypertension, diabetes, and chronic kidney disease to have the highest impact on hFwe‐Lose expression. A linear regression model was created to account the patient’s age upon analysis of hFwe‐Lose expression in relation to comorbidity status. “Other comorbidity” refers to a cumulative effect of diseases or conditions, not directly associated with COVID‐19 (cancer, Down syndrome, solid organ transplant, sickle cell disease, bone marrow transplant). The plot illustrates the effect of selected comorbidities (horizontal axis) on relative hFwe‐Lose expression (vertical axis). The P‐values of the respective linear models for all significant comorbidities are presented on the plot. The error bars represent the 95% confidence interval.

4. Elevated hFwe‐Lose expression in the nasal swab samples associates with patients’ condition severity and respective medical treatment. Box plots illustrate an increased expression of hFwe‐Lose in nasal swabs of patients, who were hospitalized within 14 days of disease progression (n = 177), admitted to intensive care unit (ICU) (n = 34), underwent intubation (n = 58), had respiratory rate greater than 30 (GT30; n = 76), had blood oxygenation level (SpO2) less than 94% (n = 147), and died within 30 days of disease progression (n = 21) versus patients without respective conditions. Pairwise two‐sided Student’s t‐tests were performed (compared to patients without respective conditions), and P‐values are presented on the plot. The vertical axis represents hFwe‐Lose expression normalized to the mean of non‐hospitalized patients. The color refers to the COVID‐19 disease outcome: gray for non‐hospitalized, blue for hospitalized, and red for deceased patients. The shape of data points reflects the cohorts: circles for the training cohort (n = 203) and triangles for the validation cohort (n = 80). The central band shows the median, the box indicates the interquartile range, and the whiskers extend to the most extreme points within the 1.5‐fold distance of the interquartile range above and below the box.

5. Elevated hFwe‐Lose expression in nasal swab associates with patients’ disease outcome. Box plot emphasizes an increased expression of hFwe‐Lose in nasal swabs of patients, who were hospitalized within 14 days of disease progression (n = 177), and who died within 30 days of disease progression (n = 21) versus patients without respective conditions. Two‐sided Student’s t‐tests were performed (compared to non‐hospitalized patients), and P‐values are presented on the plot. The vertical axis represents hFwe‐Lose expression normalized to the mean of non‐hospitalized patients. The color refers to the COVID‐19 disease outcome: gray for not hospitalized, blue for hospitalized, and red for deceased patients. The shape of data points reflects the cohorts: circles for the training cohort (n = 203) and triangles for the validation cohort (n = 80). The central band shows the median, the box indicates the interquartile range, and the whiskers extend to the most extreme points within the 1.5‐fold distance of the interquartile range above and below the box.

6. The logistic model predicts probability of hospitalization based on hFwe‐Lose expression in nasal swab samples. This model predicts a > 50% chance of hospitalization for people (otherwise still healthy, not yet infected with COVID‐19), who have a hFwe‐Lose expression in their nasal swab samples >1.82 than the mean of non‐hospitalized patients. P‐value of the logistic model < 0.001. The gray area shows the 95% confidence band.

Figure 3. hFwe‐Lose biomarker, measured in nasal swab samples, predicts patients’ COVID‐19 outcome

1. Classification and regression tree (CART) shows that hFwe‐Lose expression in patients’ nasal swab sample and patients’ age is the main predictors of the outcome. The classification tree was generated using all relevant information about patients (hFwe‐Lose expression, age, sex, presence of comorbidities (diabetes, COPD, obesity, cardiomyopathy, heart failure, hypertension)). All patients were included in the CART analysis (n = 283). The CART algorithm selected hFwe‐Lose expression and age as sole factors to determine the patients’ outcome. The split cut‐offs, which produce tree branches, are aimed to maximize the information gain (decrease of entropy) with each split, and in this way, the coefficient of determination increases, and the relative error decreases with each split (Barlin et al, 2013).

2. The line plots show saturation of the coefficient of determination, as well as the plateau in X relative error. P‐value of the classification and regression tree analysis (for all splits) < 0.01. The first split (hFwe‐Lose > 2.45) increased the coefficient of determination by ˜45% and reduced the relative error by ˜45%. The impact of all following splits on the relative error was irrelevant. The error bars (plot on the right) represent ± SE.

3. The random forest analysis shows the highest impact of the hFwe‐Lose biomarker in the multivariate analysis of outcome prediction. The plot shows the Mean Decrease in Gini coefficients for the factors that were incorporated in the statistical model for the multivariate CART analysis. The Gini coefficient is a measure of the misclassification rate. The importance of a predictor is assessed by how much the predictor reduced this misclassification rate. The hFwe‐Lose expression has the highest score, followed by age and blood biomarkers. Comorbidities show the least impact on reducing the misclassification rate.

4. hFwe‐Lose is a sensitive and specific biomarker that predicts poor COVID‐19 outcome. The ROC curves illustrate the high sensitivity and (1 ‐ specificity) of FC > 3.17 (for hospitalization; TPR = 0.77, FPR = 0.03) and FC > 4.44 (for death; TPR = 0.1, FPR = 0.08) threshold levels (AUC = 0.89 and 0.98, respectively) in prediction patients’ hospitalization and death. CI 95% (hospitalization) ‐ [0.84–0.93]; CI (death) ‐ [0.92–1]. Only retrospective (training, n = 203) patients’ cohort was used for the creation of ROC curves.

5. Elevated hFwe‐Lose expression in nasal swab probes predicts patients’ disease outcome. Box plots show an increased expression of hFwe‐Lose in nasal swabs of patients, who were hospitalized or died, versus patients who were not hospitalized, for retrospective (training; n = 203) and prospective (validation; n = 80) patients’ cohorts. Two‐sided Student’s t‐test was performed, and P‐values are presented on the plot. The vertical axis represents hFwe‐Lose expression normalized to the mean of non‐hospitalized patients. The color refers to the COVID‐19 disease outcome prediction: gray for not hospitalized, blue for hospitalized, and red for deceased patients. In the retrospective cohort, the outcome prediction was 84% correct in predicting non‐hospitalization, 63% correct in prediction of hospitalization, and 100% correct in death prediction. In the prospective cohort, the outcome prediction was 100% correct in predicting non‐hospitalization, 72% correct in prediction of hospitalization, and 55% correct in death prediction (45% of deceased patients were predicted to be “only” hospitalized; none of deceased patients had “not hospitalized” prediction). Two‐sided Student’s t‐tests were performed (compared to non‐hospitalized patients), and P‐values are presented on the plot. The central band shows the median, the box indicates the interquartile range, and the whiskers extend to the most extreme points within the 1.5‐fold distance of the interquartile range above and below the box.

6. The confusion matrices and heatmaps visualize the classification performance of hFwe‐Lose expression at the selected cut‐offs (FC > 3.17 for hospitalization and FC > 4.44 for death). In the retrospective (training) cohort, 72 out of the total 86 not hospitalized patients were correctly predicted; out of 107 hospitalized patients, 67 were correctly predicted, 19 were predicted to die instead, and 21 were not predicted to be hospitalized; all deceased patients were correctly predicted. In the prospective (validation) cohort, all patients (19) who were not hospitalized, were correctly predicted; out of 50 hospitalized patients, 36 were correctly predicted, 14 were predicted to be non‐hospitalized; out of 10 deceased patients, 6 were correctly predicted, and 5 patients were predicted to be “only” hospitalized. For hospitalization prediction, positive predictive value (PPV) for retrospective (training) cohort is 83.7% and for prospective (validation) cohort is 87.8%. The negative predictive value (NPV) for retrospective (training) cohort is 67.2% and for prospective (validation) cohort is 64.1%. For death prediction, positive predictive value (PPV) for retrospective (training) cohort is 34.5% and for prospective (validation) cohort is 100%. The negative predictive value (NPV) for retrospective (training) cohort is 100% and for prospective (validation) cohort is 93.2%.

Figure 4. The linear regression models show superiority of the hFwe‐Lose biomarker to predict COVID‐19 outcome, compared with conventional biomarkers

1. hFwe‐Lose predicts COVID‐19 patients’ death more accurately than other biomarkers. hFwe‐Lose was compared with four known biomarkers (ferritin, CRP, D‐dimer, and neutrophil‐lymphocyte ratio, respectively) in predicting death of hospitalized patients. For each biomarker, only patients who had the information of the respective blood biomarker were used. 115 patients were used to compare ferritin and hFwe‐Lose. 120 patients were used to compare D‐dimer and hFwe‐Lose. 127 patients were used to compare CRP and hFwe‐Lose. 153 patients were used to compare neutrophil–lymphocyte ratio and hFwe‐Lose. With AUC as the criteria, hFwe‐Lose significantly outperformed all four biomarkers in predicting death in both retrospective and prospective cohorts. AUC coefficients, as well as CIs, are displayed on the plots.

2. hFwe‐Lose biomarker is superior to other markers in COVID‐19 poor outcome prediction. All 283 patients were used to compare hFwe‐Lose and age combined with comorbidities in predicting hospitalization (left) and death (middle). The AUC of age combined with comorbidities in predicting the hospitalization of the prospective cohort is 0.88 (CI ‐ [0.81–0.96]). The AUC of hFwe‐Lose in predicting the hospitalization of the prospective cohort is 0.90 (CI ‐ [0.84–0.97]). hFwe‐Lose outperformed age combined with comorbidities in predicting hospitalization. The AUC of age combined with comorbidities in predicting the death of the prospective cohort is 0.86 (CI ‐ [0.72–1.0]). The AUC of hFwe‐Lose in predicting the death of the prospective cohort is 0.98 (CI ‐ [0.92–1.0]). hFwe‐Lose significantly outperformed age combined with comorbidities in predicting death. The 105 patients who registered the information of all four known blood biomarkers (ferritin, CRP, D‐dimer, and neutrophil‐lymphocyte ratio) were used to compare hFwe‐Lose, age combined with comorbidities, and the four biomarkers in predicting death (right). hFwe‐Lose significantly outperformed age combined with comorbidities and the four biomarkers in predicting death in both retrospective and prospective cohorts derived from the 105 patients.