Unbiased plasma profiling using pre-selected RNA aptamer pools predicts mortality in COVID-19 and identifies protein risk factors

Abstract

The impact of the COVID-19 pandemic demands effective prognostic tools for precise risk evaluation and timely intervention. This study utilized the APTASHAPE technology to profile plasma proteins in COVID-19 patient samples. Employing a highly diverse 2′-fluoro-protected RNA aptamer pool enriched toward proteins in the plasma samples from COVID-19 patients, we performed a single round of parallel selection on the derivation cohort and identified 93 discriminatory aptamers capable of distinguishing COVID-19 and healthy plasma samples. A subset of these aptamers was then used to predict 30-day mortality with high sensitivity and specificity in a validation cohort of 165 patients. We predicted 30-day mortality with areas under the curve (AUCs) of 0.91 in females and 0.68 in males. Affinity purification coupled with mass spectrometry analysis of the aptamer-targeted proteins identified potential biomarkers associated with disease severity, including complement system components. The study demonstrates the APTASHAPE technology as an unbiased approach that not only aids in predicting disease outcomes but also offers insights into gender-specific differences, shedding light on the nuanced aspects of COVID-19 pathophysiology. In conclusion, the findings highlight the promise of APTASHAPE as a valuable tool for estimating risk factors in COVID-19 patients and enabling stratification for personalized treatment management.

Graphical abstract

Kjems and colleagues used the APTASHAPE technology to profile plasma proteins, identifying 93 aptamers distinguishing COVID-19 and control patients. The aptamers upregulated in COVID-19 predicted 30-day mortality with high accuracy. This proteomic approach reveals potential biomarkers linked to disease severity, emphasizing APTASHAPE’s potential to stratify patients and guide treatment decisions.

Introduction

Coronavirus disease 2019 (COVID-19), caused by infection with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has tremendously impacted our societies and overwhelmed healthcare systems worldwide.¹ Public vaccination programs and the SARS-CoV-2 infection itself have led to broad immunity in most populations, and many countries have been without COVID-19 restrictions for the past year. However, the emergence of new SARS-CoV-2 variants like BA.2.86 (summer 2023), which carries 34 mutations in the spike protein compared to the BA.2 virus and 58 mutations compared to the original Wuhan index virus, has once again raised awareness of monitoring COVID-19 closely.^2,3,4

The symptoms of COVID-19 are widespread, ranging from common symptoms like dry cough, fever, headache, sore throat, dyspnea, or myalgia⁵ to symptoms of more severe COVID-19 manifestations like chest pain, pneumonia, hypoxemia, confusion, and other complications requiring intensive care unit (ICU) admission.⁶ The severity and mortality of COVID-19 infections have been linked to different risk factors.^7,8,9,10 These include gender, age, and comorbidities, such as hypertension, cancer, cardiovascular diseases, obesity, diabetes, and asthma. Also, pathogenic genetic variants have been reported to disrupt immune homeostasis, potentially causing inadequate immune responses to viral infections or overproduction of pro-inflammatory cytokines.^11,12 Yet, individuals without comorbidities may also experience severe symptoms, warranting the importance of adequate prognostic and diagnostic tools for COVID-19.

Generally, biomarkers are used as tools to stratify the prognosis of COVID-19 outcomes and determine treatment plans. Several biomarkers, such as white blood cell count; decreased lymphocyte and platelet counts; elevated plasma levels of creatinine kinase, alanine aminotransferase, lactate dehydrogenase, C-reactive protein, ferritin, interleukin-6, PTX3, and D-dimer; and signs of complement aberrations, have been associated with severe illness and mortality in patients with COVID-19.^{13,14,15,16,17,18}

Here, we aimed to identify biomarkers associated with COVID-19 in an unbiased fashion and independent of co-existing medical conditions using APTASHAPE technology. In a previous study, this technology was able to reliably diagnose different stages of bladder cancer progression with high accuracy.¹⁹ Based on this, we hypothesized that APTASHAPE technology can also act as a high-throughput platform to predict COVID-19 disease outcome and identify COVID-19 biomarkers.

We report here that APTASHAPE technology can address the protein composition of COVID-19 plasma samples using serum-stable oligonucleotide aptamers. Changes in the ratio of these 2′-fluoro-protected RNA sequences (aptamers) are used to decipher changes in the protein compositions of blood plasma.^19,20 Thus, protein composition can be translated into sample-specific patterns of aptamers that are readily profiled by next-generation sequencing (NGS) and machine learning algorithms.

Initially, we applied a small COVID-19 cohort to identify aptamers able to distinguish the plasma protein signature of healthy controls and COVID-19 patients. Using a validation cohort, we identified a set of discriminatory aptamers that could predict 30-day mortality after admission to the hospital with high sensitivity and specificity in the large validation cohort (n = 165), although this was most significant for females. The most abundant members of each aptamer family (the predictive aptamers) were subsequently synthesized individually and used as bait for target protein purification. Mass spectrometry (MS) analysis revealed that a large proportion of the targeted proteins are members of the complement system and immune-related pathways, which is in accordance with our current understanding of COVID-19 disease manifestation. The study demonstrates APTASHAPE technology’s ability to estimate risk factors for COVID-19 patients and stratify patients based on aptamer profiling of plasma samples.

Results

Enrichment of the RNA sequence pool toward COVID-19-specific characteristics

An initial pool of 10E15 RNA sequences, consisting of a randomized region of 36 nucleotides flanked by two constant regions facilitating RT-PCR amplification, was enriched for RNA sequences (aptamers) able to bind proteins from whole plasma mixed from 5 hospitalized COVID-19-positive patients. This was accomplished by first immobilizing the pooled plasma proteins on NHS-activated magnetic beads, followed by four iterative rounds of systematic evolution of ligands by exponential enrichment (SELEX). Here, the binding specificity of the aptamer pool targeting COVID-19-specific signatures was enhanced by performing negative selection toward whole plasma from non-infected controls (Figure 1A). After each round of SELEX, the enrichment of target binding aptamers was determined by RT-PCR analysis of bead-bound RNA and pool diversity examined by NGS (data not shown). Based on these results, the aptamer pool after 4 rounds of selection was chosen as most appropriate for multiplex profiling of individual patient samples (branched selection). At this point, the aptamer pool’s diversity was still large, showing 85,912 different aptamer sequences. Similarity clustering was performed to remove near-identical sequences, which likely arise from the same ancestor clone and therefore exhibit affinity for the same protein. This reduced the pool to 32,492 aptamers (Levenshtein-Damerau distance ≤ 3), which varied in abundance between 0.00002 and 53 percent (Figure S1).

Figure 1.

Schematic representation of the APTASHAPE technology

(A) The random RNA sequence pools were enriched toward COVID-19-specific characteristics through systematic evolution of ligand by exponential enrichment (SELEX). Iterative rounds of negative and positive selections were performed against non-COVID-19 plasma (healthy controls) and COVID-19 plasma, respectively. RNA sequences with an affinity toward the proteins in the sample of interest (COVID-19 plasma) were recovered and subjected to cDNA synthesis, PCR amplification, and in vitro transcription, regenerating the enriched RNA sequence pool for the following selection round. (B) Illustration of the branched selection concept. An enriched RNA sequence pool is mixed with samples of interest in a parallelized fashion (branched selection). The branched selections led to aptamer binding profiles representing sample-specific fingerprint-like imprints of each sample. Disease-specific patterns are identified by comparing binding profiles across sample subgroups using machine learning algorithms. Aptamers capable of distinguishing sample subgroups were identified, and the signals from these discriminative RNA aptamers were used to predict COVID-19 30-day mortality.

APTASHAPE profiling of COVID-19 patients' plasma proteins

The fourth-round RNA pool was then subjected to one round of selection toward individual COVID-19 plasma samples (branched selection, Figure 1B). This derivation cohort included four hospitalized COVID-19 patients and four healthy controls. In two technical replicates, aptamer binders were isolated, identified, and quantified using NGS, translating the individual sample-specific binding profiles to quantitative data. A mean of 5.6 million sequences per sample across all individual samples was obtained, which led to the recognition of all 85,912 different aptamer sequences. To enhance the statistical robustness of the dataset, we performed similarity clustering (Levenshtein-Damerau distance ≤ 3) and set a threshold requirement that each aptamer must be present more than four times across all samples (total count). This reduced the number of aptamers to 5,313 sequences. The abundance of the 5,313 sequences was normalized to the percentage of total reads to allow for the comparison of aptamer levels between samples. Furthermore, subgroup-specific trends in aptamer levels were identified by dividing individual aptamer reads with the mean read number of the respective aptamer.

To eliminate potential bias introduced by the absence of some aptamers in a subset of the samples, i.e., zero values, due to lack of measurement sensitivity, we performed a discriminative analysis of subgroup-specific trends on the top 2,000 most abundant aptamers, which are highly represented across all samples; Figure S2). The nucleotide composition and sample-specific counts of these 2,000 aptamers are provided in Table S1.

RNA aptamers able to distinguish the COVID-19 and healthy control samples were found by ordinary least squares linear regression using the healthy controls as the reference. We identified 93 significantly discriminating aptamers exhibiting a change in subgroup abundance ratio (regression coefficient) of >0.75 and Benjamini-Hochberg adjusted p value of <0.05 (Figure 2, blue and green dots; Table S2). Of these, 19 aptamers had a higher relative abundance mean (Figure 2, green dots) and 74 aptamers had a lower relative abundance mean (Figure 2, blue dots) in the COVID-19 subgroup compared to the healthy subgroup.

Figure 2.

Identification of aptamers discriminating COVID-19 and non-COVID-19 patient samples

Ordinary least square (OLS) linear regression analysis was performed with a regression coefficient at ≥0.75 and a Benjamini-Hochberg-adjusted significant level at 0.05 to identify COVID-19 discriminative RNA aptamers. The volcano plot shows the relative abundance means, i.e., up- and downregulation of the aptamers, and aptamers meeting the predefined discrimination ability are indicated by green and blue dots, respectively. The analysis revealed 93 RNA aptamers discriminating non-COVID-19 controls and COVID-19 patient plasma samples (blue and green dots). The aptamers included in the mass spectrometry (MS) analysis are indicated by a red outline and shown in Table 2.

The discriminatory aptamer sequences (Table S2) were subjected to unbiased homology analysis, which identified 29 unique aptamer families based on conserved sequence motifs (color coded in Table S2). Of these families, 7 were found within the upregulated aptamers, while 22 were found in the downregulated aptamers. Aptamers within the same family are expected to bind similar epitopes in the proteins, and the identification of 29 families suggests that the discriminatory aptamers recognize a panel of different targets in the plasma samples.

Validation of the discriminative aptamers for prediction of COVID-19 outcome

We hypothesized that aptamers with the ability to detect protein composition changes in plasma from COVID-19 patients could predict the severity of disease development. To test this, plasma samples were collected from 165 COVID-19 patients (113 males and 52 females) within the first 4 days of admission to Copenhagen University Hospital - Amager and Hvidovre, Denmark (see Table 1 for baseline characteristics of the cohort), and branched selections were performed.

Table 1.

Description of the validation cohort

Variables	All (n = 165)	Survivors (n = 141)	Non-survivors (n = 24)	p value
Age group, years
<50	40 (24.2)	40 (28.4)	0 (0)	–
50–59	31 (18.8)	31 (22)	0 (0)	–
60–69	35 (21.2)	28 (19.9)	7 (29.2)	–
70–79	37 (22.4)	24 (17)	13 (54.2)	–
≥80	22 (13.3)	18 (12.8)	4 (16.7)	–
Female	52 (31.5)	45 (31.9)	7 (29.2)	–
Treatment
Remdesivir	141 (85.5)	121 (85.8)	20 (83.3)	0.75
Dexamethasone	150 (90.9)	128 (90.8)	22 (91.7)	1
Comorbidity
Arterial hypertension	55 (33.3)	45 (31.9)	10 (41.7)	0.36
Diabetes	46 (27.9)	35 (24.8)	11 (45.8)	0.048
Cardiovascular disease	62 (37.6)	45 (31.9)	17 (70.8)	0.00046
Chronic obstructive pulmonary disease	12 (7.3)	9 (6.4)	3 (12.5)	0.3854
Malignant cancer	24 (14.5)	16 (11.3)	8 (33.3)	0.0098
Asthma	22 (13.3)	20 (14.2)	2 (8.3)	0.7445
Other diseases	115 (69.7)	93 (66)	22 (91.7)	0.01446
Clinical presentation
Body mass index	28.35 kg/m2 (24.97, 32.08)	28.36 kg/m2 (25.28, 32.05)	26.88 kg/m2 (24.58, 32.84)	0.81
Lung infiltrates at baseline	148 (89.7)	126 (89.4)	22 (91.7)	1
Days from the first symptom to admission	8 days (5, 9)	8 days (5, 9)	5.5 days (3, 6.75)	0.0098
Respiratory rate	23 (20, 28)	22 (22, 26)	25 (20, 32.5)	0.079
Oxygen baseline
No oxygen	11 (6.7)	10 (7.1)	1 (4.2)	1
Low flow	78 (47.3)	75 (53.2)	3 (12.5)	0.00025
High flow	73 (44.2)	54 (38.3)	19 (79.2)	0.00025
Mechanical ventilation	3 (1.8)	2 (1.4)	1 (4.2)	0.38

Values denote the median (interquartile range) or number (%). p values are Fisher’s exact test.

The distribution of aptamers in the pool obtained after branched selections was profiled by NGS as described for the derivation cohort (Table S3). We found all 93 discriminatory aptamers from the derivation cohort to be present in the dataset after branched selections. We suspected the previously identified 19 upregulated discriminatory aptamers (Figure 2, green dots) to be predictive of disease severity and evaluated the performance of this potentially predictive panel in the validation cohort.

As a measure of disease severity, we chose the 30-day mortality calculated from the admission date. The ability of the panel to distinguish 30-day survival vs. non-survival was evaluated using principal-component analysis (PCA; Figure 3A). While the first PC (PC1 34.8%) separated gender (Figure S3), PC2 explains 15.47% of the variation in the dataset and separated 30-day survival vs. non-survival. A non-parametric Wilcoxon rank-sum test was performed to evaluate the probability of the result, which showed that the difference in the median value between the survival vs. non-survival subgroups is statistically significant with a p value of 0.0061. Due to the observed gender effect and the reported gender effect on average disease severity,²¹ the validation cohort was stratified to separate the genders, and PCA was performed on the gender-specific cohorts.

Figure 3.

Validation of the discriminatory aptamers' capability to predict 30-day mortality

Blood samples collected from COVID-19 patients upon admission to the ICU were analyzed using the discriminatory aptamers upregulated in COVID-19, which the OLS analysis identified in the derivation cohort. The potential of the discriminatory aptamers to predict 30-day mortality as a means of disease severity was evaluated by principal-component analysis (PCA). PCA was performed due to its ability to reduce the dataset’s complexity while retaining most of the information. (A) The boxplot shows the PC2 values and 30-day mortality (0 = survival, 1 = non-survival) for the validation cohort. The difference in median values between survival and non-survival was statistically significant, with a Wilcoxon rank-sum test p value of 0.0061. (B) PCA of the female subpopulation of the validation cohort. The boxplot shows the PC1 values and 30-day mortality (0 = survival, 1 = non-survival). A significant separation is observed between survival and non-survival, with a p value of 0.00015. (C) PCA of the male subpopulation of the validation cohort. The boxplot shows the PC2 values and 30-day mortality (0 = survival, 1 = non-survival). A significant separation is observed between survival and non-survival, with a p value of 0.02. (D) AUCROC curve for 30-day mortality (survival and non-survival) for the data shown in (A)–(C). The selected aptamers discriminate survival and non-survival with high sensitivity and specificity for the female subpopulation (B) with an AUROC of 0.914. The data in (A) and (C) have AUCROC values of 0.676 and 0.678, respectively.

For the female cohort, PC1 now clearly separated the survival vs. non-survival subgroups, accounting for 33.9% of the variation in the data (Figure 3B). A Wilcoxon rank-sum test found that the difference in median values was statistically significant with a p value of 0.00015. The separation of the male subpopulation was less clear, exhibiting a p value of 0.02 (Figure 3C). The overall performance of the panel in all subpopulations was evaluated using the area under the receiver operating characteristic curve (AUROC) analysis. The probability curve and the measure of separability are shown in Figure 3D. The panel distinguishes 30-day survival and non-survival in females and males with an AURUC of 0.91 (95% confidence interval [CI]: 0.8152–1, DeLong’s test) and 0.68 (95% CI: 0.5053–0.8501, DeLong’s test), respectively. We conclude that the panel of 19 aptamers with a relatively higher abundance in COVID-19 patients identified in the derivation cohort could predict disease severity with high accuracy in females but to a lesser extent in the male population.

Identification of target proteins

To identify the corresponding protein biomarkers predictive of disease severity and diagnosis of COVID-19, affinity purification of plasma proteins was performed using the most abundant aptamer sequence within each aptamer family as bait. The 30 selected aptamers (Table 2; Figure 2, dots with red outline) were grouped according to relative abundance in selections from COVID-19 patients, i.e., 7 upregulated aptamer families and 22 downregulated aptamer families. Furthermore, the selected aptamers were grouped into four main subgroups according to the APTASHAPE binding profiles by hierarchical cluster analysis to investigate if aptamers with analogous binding profiles share similar epitopes to those expected for aptamers within the same family (Figure 4A).

Table 2.

Overview of aptamers selected for affinity purification of protein targets

Name	Sequence, 5′ → 3′	Family	Relative abundance in COVID-19 patients	Suggested candidate biomarkers (% of total sample intensity)	Suggested candidate biomarkers (% of total sample intensity)	Suggested candidate biomarkers (% of total sample intensity)
AptCov101	TGCGGTTTATGAGCAATCATATCGCTCGCGTTGTCA	W	lower in COVID-19	DCD P81605 (11)	CALL5 Q9NZT1 (6.78)	IGKC P01834 (6.25)
AptCov102	AATCCAATACCGGCGAGCCGAAATCCTTGCCAACGG	M	lower in COVID-19	HPT P00738 (13.44)	ITIH2 P19823 (10.23)	IGKC P01834 (5.98)
AptCov103	TTCGCATGCTAAACGATGCAACTTGTAGCTGAGTTG	K	lower in COVID-19	ITIH2 P19823 (14.05)	HPT P00738 (11.3)	ITIH1 P19827 (6.53)
AptCov104	TGCACGCGACTGGTCGTAACCAGTCCATTTGTCATG	T	lower in COVID-19	HPT P00738 (16.84)	DCD P81605 (8.11)	ITIH2 P19823 (7.59)
AptCov105	GTCGCGCAAGTCACTTCGCGCCTTGTAGCTGTGTTG	A	lower in COVID-19	HPT P00738 (17.91)	ITIH2 P19823 (14.01)	ITIH1 P19827 (6.48)
AptCov106	CATGGTTGTGAATCCACAGAGATTCGATGTTTGCGC	V	lower in COVID-19	ITIH2 P19823 (12.97)	HPT P00738 (7.15)	ITIH1 P19827 (6.32)
AptCov107	TATGGCTGTGATGCTAGTTCAGCATCGATACTCGTC	Q	lower in COVID-19	HPT P00738 (15.59)	ITIH2 P19823 (13.07)	ITIH1 P19827 (5.99)
AptCov108	CGACCCTCCGCGTAAGCGGCGTCGTTGTCAGATGTA	P	lower in COVID-19	ITIH2 P19823 (11.98)	HPT P00738 (8.19)	DCD P81605 (5.8)
AptCov109	GTGACCGTCCAGGCCACATCGTCCCTTGCGGGTATC	I	lower in COVID-19	HPT P00738 (16.51)	ITIH2 P19823 (10.98)	ITIH1 P19827 (5.31)
AptCov110	ACGGCTGAGGAAGTTAACGTACTTCTGGTATCGTTC	J	lower in COVID-19	ITIH2 P19823 (15.55)	HPT P00738 (13.6)	ITIH1 P19827 (7.49)
AptCov111	CAGCCATGACTTGATTCACATCCTCATGGTCCTTGT	G	lower in COVID-19	HPT P00738 (16.76)	ITIH2 P19823 (8.06)	IGKC P01834 (6.59)
AptCov112	CCGTTATGATTTGATTTACATCCTCATAGCTAGTGT	H	lower in COVID-19	ITIH2 P19823 (24.2)	ITIH1 P19827 (9.97)	AMBP P02760 (4.67)
AptCov113	TGCGAAAACTCGCATTGATGGAGTATAACTACCTTC	L	lower in COVID-19	ITIH2 P19823 (21.77)	ITIH1 P19827 (11.24)	HPT P00738 (11.06)
AptCov114	AAGTGTCATGATTTGACTTTCATCCTCATGACTTTC	CC	lower in COVID-19	ITIH2 P19823 (11.64)	HPT P00738 (8.35)	IGKC P01834 (5.45)
AptCov115	CACAGATGTACACCCTCACGGTGTGGTGCTTAGATC	U	lower in COVID-19	HPT P00738 (18.28)	ITIH2 P19823 (10.84)	IGKC P01834 (5.6)
AptCov116	GGGCGTCAACTTGCACCGAAATCTGCAGCCTCAATC	BB	lower in COVID-19	HPT P00738 (11.4)	ITIH2 P19823 (9.72)	IGKC P01834 (5.98)
AptCov117	CATAGTTGTAGCGGTCAGTGATCGCTGATGTCCGTC	E	lower in COVID-19	ITIH2 P19823 (18.11)	HPT P00738 (9.08)	ITIH1 P19827 (6.84)
AptCov118	ACAGCTGAGGCAGCTAGCTAACTGCCGGTATCGATC	C	lower in COVID-19	HPT P00738 (16.82)	ITIH2 P19823 (11.54)	ITIH1 P19827 (4.33)
AptCov119	AGCACGAACAATCAAAGATGTACTGCTCCTCGCAGT	F	lower in COVID-19	HPT P00738 (20.63)	ITIH2 P19823 (9.93)	IGKC P01834 (5.94)
AptCov120	TGTCCGTACTGACTTCGTACGGCACTTGTAGCATGT	X	lower in COVID-19	C4BPA P04003 (6.23)	IGKC P01834 (6.05)	HPT P00738 (4.81)
AptCov121	CACATGCTACCCGCTTAGCGAATGATAGCGGGACTC	AA	lower in COVID-19	HPT P00738 (14.68)	ITIH2 P19823 (9)	IGKC P01834 (5.53)
AptCov122	AACCGTGTATCCACTTGACGTGCCTTACTAGGTGGC	D	lower in COVID-19	HPT P00738 (9.99)	ITIH2 P19823 (6.14)	IGKC P01834 (5.8)
AptCov123	AATGCACAGTGGCGCATCCATCTGAGATACCGGATG	B	higher in COVID-19	HPT P00738 (7.85)	CALL5 Q9NZT1 (6.92)	IGKC P01834 (6.44)
AptCov124	GTAGGTAGACAACGGTAGTGCGTTGTAACATTCATC	R	lower in COVID-19	ITIH2 P19823 (9.09)	CO3 P01024 (4.44)	HPT P00738 (3.29)
AptCov125	TGAAAATCGTAACACAGTATGGACCTGTGCAATCCC	O	higher in COVID-19	HPT P00738 (26.89)	IGKC P01834 (7.15)	TRFE P02787 (6.52)
AptCov126	GAGCACCAGCTGACCACGGCGTGGCCTACCGCTTCA	PP	higher in COVID-19	HPT P00738 (15.2)	IGKC P01834 (6.72)	C4BPA P04003 (3.95)
AptCov127	CGGTGCGCTAGGCCGGAGGAACACGTCTACTCCATC	DD	higher in COVID-19	HPT P00738 (17.35)	IGKC P01834 (6.23)	IGHG3 P01860 (4.19)
AptCov128	CTATGGACATTTCAACGCCGTCGTATCGGCTCACTC	X	higher in COVID-19	HPT P00738 (14.98)	IGKC P01834 (6.75)	IGHG3 P01860 (3.69)
AptCov129	GTTCGATCACAACGAACTGATTGTGCGACCCGTAAC	X	higher in COVID-19	CFAH P08603 (67.08)	HPT P00738 (2.85)	IGKC P01834 (2.54)
AptCov130	GTTGTATTGACACGCACGGTGTCAGCACCCTTGACC	S	higher in COVID-19	CFAH P08603 (77.08)	HPT P00738 (4.96)	IGKC P01834 (1.43)

Enriched sequence motifs for each of the 30 discriminatory aptamers are colored. Aptamer families are identified by "LETTERS" and relative expression in the derivation cohort. The affinity-purified plasma proteins were analyzed by LC-MS/MS, and the most abundant targets (UniProtKB IDs) for each aptamer are shown (percentage of total protein signal intensity). For detailed information, see Figure 4B.

Figure 4.

Aptamer target identification

The most abundant members of each aptamer family were selected for protein target identification to identify potential disease-associated biomarkers. (A) The selected aptamers (Apt101–130) were grouped using hierarchical clustering on their binding profiles (relative abundance) in COVID-19 samples, leading to the identification of 4 general APTASHAPE binding patterns (subgroups G1–G4) among the selected aptamers. (B) Affinity purification using Apt101–130 was performed on mixed plasma samples from COVID-19 patients. The purified proteins were analyzed by MS (LC-MS/MS), identified and quantified using Proteome Discoverer label-free quantitation (LFQ), and searched against the human database from UniProt. The heatmap shows the 25 most abundant proteins recovered in the affinity purification across all samples (Apt101–130) and the log2-transformed sample percentage for each sample. The most enriched proteins are shown in dark red. The affinity purification was done under native conditions, allowing the purification of proteins sitting in protein complexes.

Recovered proteins were analyzed by MS, identified and quantified using Proteome Discoverer label-free quantitation (LFQ), and searched against the human database from UniProt.²² The dataset contains signals for 169 unique proteins, setting the Score SequestHT threshold to ≥2. To compare samples, the percentage of the total signal intensity for each protein was calculated for each sample, log2 transformed, and visualized using hierarchical cluster analysis (Figure 4B; most abundant targets for each aptamer are summarized in Table 2). All aptamers, including the unrelated control aptamers, yielded a strong signal for the most abundant plasma protein, serum albumin, as well as various keratin proteins, which we assume are contaminants, as previously reported by Fjelstrup et al.¹⁹ We also observed strong signals for serum amyloid (UniProtKB P02743), immunoglobulin heavy constant mu (UniProtKB P01871), and immunoglobulin kappa constant (UniProtKB P01831) across all samples, indicating background binding.

More specifically, aptamers 102–120 + 122 + 124, which have a relatively low abundance in COVID-19 samples, yielded strong signals for inter-alpha-trypsin inhibitor heavy chain H2 (UniProtKB P19823), inter-alpha-trypsin inhibitor heavy chain H1 (UniProtKB P19827), and protein AMBP (UniProtKB P02760). In addition, the aptamers 123 + 125–128, which were upregulated in COVID-19, yielded a more consistent signal for serotransferrin (UniProtKB P02787), immunoglobulin gamma-1 heavy chain (UniProtKB P0DOX5), immunoglobulin heavy constant gamma 3 (UniProtKB P01860), and immunoglobulin lambda-like polypeptide 5 (UniProtKB B9A064) than the downregulated ones. Furthermore, a strong signal of complement factor H was observed for aptamers 129 and 130, which are upregulated in the COVID-19 subpopulation of the derivation cohort.

Discussion

Accumulating data demonstrate that variations in the plasma protein composition mirror our health status and that the plasma protein composition can help understand disease etiology.²³ The ability to globally profile the plasma proteomic landscape enables the identification of protein signatures associated with pathological disease and paves the way for developing diagnostic and prognostic assays. The APTASHAPE technology deciphers protein signatures using aptamers and, as demonstrated here, enables stratification of the high-risk COVID-19 patients at an early phase of the disease. The massive impact of the COVID-19 pandemic on global health has accelerated the identification of prognostic biomarkers, including long pentraxin PTX3,¹⁸ immune signatures,^11,24 and well-established laboratory biomarkers and demographic information.^25,26 Our APTASHAPE approach is an NGS-based analysis and does not impose any bias or requirement for prior knowledge about the protein content and patient to detect fluctuations in proteins in the plasma from COVID-19 patients.

Furthermore, the aptamers are expected to interact with specific sites in the proteins, and we have previously demonstrated that aptamer libraries, in addition to the protein concentration, can detect single-amino-acid substitution, allowing the identification of rare genetic biomarkers as well as post-translation protein modifications and conformers.²⁰ In line with our previous observation that APTASHAPE is capable of stratifying bladder cancer stages and healthy controls,¹⁹ we here provide evidence that it also can predict disease severity and outcome. In the derivation cohort, we identified 93 aptamers, representing 29 aptamer families, that could separate COVID-19-positive plasma from negative plasma. Seven aptamer families were found to be more abundant in COVID-19-positive patients, and we hypothesized that the signal from these seven aptamer families could be predictive of disease severity. Using the 30-day mortality endpoint, the seven aptamer families exhibited AUROCs of 0.91 and 0.68 for the female and male subpopulations, respectively. The survival prediction in the female subpopulation is found on PC1, indicating a uniform progression of disease traits and a homogeneous population of patients. The separation in the male subpopulation is distinct but most pronounced in PC2, suggesting that other traits in this population overshadow the state of COVID-19 disease. Interestingly, a strong gender effect has previously been noted in COVID-19 severity. Males are more vulnerable to complications from COVID-19 than females, and current data imply that health status, including comorbidities and lifestyle (smoking, use of alcohol, etc.), also creates biases in COVID-19 severeness.⁹

We performed liquid chromatography-tandem MS (LC-MS/MS) analysis of the aptamer affinity-purified plasma proteins to provide new insight into the disease etiology and identify candidate biomarkers. Since many plasma proteins are part of large protein complexes, and the affinity purifications are conducted under non-denaturing conditions, it is difficult to identify specific protein targets. We also cannot rule out that aptamers selected using our approach can bind more than one target. The aptamers 129–130, with a relatively higher abundance mean in COVID-19 patients, show particularly strong signals for complement factor H, which regulates the alternative pathway in the complement system. Complement dysregulation has been linked to severe disease progression through SARS-CoV-2 spike protein blockage of the complement factor H binding site on heparin.²⁷ Furthermore, single-nucleotide variants in complement factor genes, including complement factor H, have been identified as risk factors for morbidity and mortality in COVID-19.²⁸ Our findings show that complement factor H is central to the ability of our aptamer panel to predict disease severity, consistent with its crucial role in regulating the activity of the complement system.²⁹ Complement factor H and complement factor H-like proteins downregulate the alternative pathway, while the 5 complement factor H-related proteins exert an opposite function by promoting the activity of the alternative pathway.²⁹ Complement dysregulation is reported to be central in COVID-19 pathophysiology.^30,31 In line with this, the plasma concentration of factor H is reported to be increased by 2-fold in ICU patients compared to non-ICU patients.³² In addition to complement factor H, the factor H protein family consists of 6 other highly homologous proteins controlling the complement alternative pathway associated with susceptibility to various diseases.

The aptamers 125–128, which also have a relatively higher abundance mean in COVID-19 patients, present signals from immunoglobulin gamma heavy chains (IGG1 and IGHG3), immunoglobulin lambda-like polypeptide 5 (IGLL5), and serotransferrin. A proteomic MS analysis of serum from COVID-19 patients with a mild disease course reported IGG1 and serotransferrin to be upregulated.³³

Our data suggest that the signature of these proteins, reflected by the RNA aptamers' binding profiles, can be used to stratify at-risk patients early in the disease. The aptamer libraries are selected under native conditions, enabling the APTASHAPE libraries to recognize changes in protein conformations, post-translational modifications, multimerization, and engagement with other binding partners. The affinity purification was performed under similar conditions, meaning that proteins will be in their native conformation and likely situated in larger protein complexes. Further investigation is required to identify the epitope for our aptamers and evaluate the exclusiveness and robustness of these protein biomarkers.

The aptamers 102–119 + 121-122 + 124, which have a relatively lower abundance mean in COVID-19, signals include inter-alpha-trypsin inhibitor heavy chain H1 (ITIH1), inter-alpha-trypsin inhibitor heavy chain H2 (ITIH2), and protein AMBP. Proteomic profiling of serum from COVID-19 patients found that these proteins are consistently more abundant in survivors than in non-survivors, and a positive correlation was found in the expression of ITIH1 and ITIH2.³⁴ Our data indicate that low levels of these proteins are associated with fatal disease course. The aptamers, which have a relatively lower abundance mean in COVID-19, are essential in separating COVID-19 from healthy controls but were not included in the panel used to predict 30-day mortality and therefore only provide information about COVID-19 in relation to COVID-19-negative individuals.

The aptamers derived from the SELEX selection toward complex biofluids are biased toward the most abundant proteins due to the inherent properties of the random immobilization of proteins on the solid support matrix that enables partition of the aptamer expression affinity for the samples.^35,36 This is also a limitation of the APTASHAPE technology, and we may lose aptamers to less abundant proteins. The method could be improved by depletion of the abundant proteins or the most abundant aptamers. However, by only performing four rounds of SELEX selection, the RNA aptamer library retains a high RNA aptamer diversity, enabling the profiling of a broad range of proteins. The disease-specific aptamers are also identified according to differential abundance in the cohort and not the overall abundance.¹⁹

Furthermore, RNA aptamers recognize their protein target by a highly variable content of ionic, van der Waals, and hydrogen-bonding interactions, and aptamers will be biased toward proteins with a natural affinity for nucleic acid, in particular regions that are positively charged.³⁷ Also, some proteins may reside inside larger complexes and therefore not be assessable to aptamer binding. Consequently, the targeted proteins addressed by the APTASHAPE aptamer library may be limited to a subset of the plasma proteins.

This study demonstrates the feasibility of selecting a panel of aptamers with the capacity to diagnose COVID-19 and predict disease mortality at the time of admission to the ICU. The data suggest that the APTASHAPE technology holds great promise in estimating risk factors for COVID-19 patients and enables stratification of the patients in terms of treatment management early in the course of the disease.

Materials and methods

Patient samples

The derivation cohort included adults 18 years or older admitted to Copenhagen University Hospital - Amager and Hvidovre with confirmed SARS-CoV-2 infection between March 10 and May 31, 2020, as previously described.³⁸ The derivation cohort included 4 COVID-19-positive and 4 COVID-19-negative patients. The cohort was used to identify aptamers discriminating between COVID-19 positive and COVID-19 negative. An aptamer panel for predicting disease severity was built using the aptamers upregulated in COVID-19. The validation cohort included adults (n = 165) admitted to Copenhagen University Hospital - Amager and Hvidovre between September 7 and December 14, 2020. The patients were administered remdesivir and dexamethasone before blood sampling, while the patients in the derivation cohort did not receive this treatment. The cohorts have been described previously.³⁹ The study was approved by the Regional Data Protection Center (P-2020-262), the Danish Patient Safety Authority (record no. 31-1521-309), and by the Regional Committee on Health Research Ethics (H-20040649). The requirement of individual informed consent was waived by the committee.

A blood sample was drawn within 4 days of admission. Plasma was produced by centrifugation, and samples were stored at −80°C.

SELEX protocol

The preparation of the random RNA sequence input pools, the SELEX procedure, and branched selections were performed as previously described¹⁹ with minor modifications. To guide the aptamer pools toward COVID-19-specific characteristics, 1 μL of pooled COVID-19 positive plasma was immobilized on 1 mg NHS-activated magnetic beads in the positive selection step, and binding aptamers were partitioned. To enhance COVID-19 characteristics and remove matrix binding aptamers, 1 μL of pooled non-COVID-19 plasma (healthy controls) was immobilized on 1 mg of NHS-activated magnetic beads in the negative selection step, and the non-binding aptamer was incubated with the positive selection matrix. Based on the sequence diversity, the RNA pool from the fourth selection round was selected for branched selections. The RNA sequence pools recovered from each sample after branched selections were prepared for NGS using forward and reverse PCR primers containing sample-specific barcodes, Illumina NGS adaptors sequences, and Illumina NGS primer sequences, allowing multiplexing. The branched selection samples were pooled equimolar, and the pooled sample was submitted to the NGS core facility at Aarhus University Hospital, Aarhus University, and sequenced on the Illumina NovaSeq machine (NovaSeq S1, 300 [400–500 Gb, 1.3–1.6 B reads] v.1.5) according to the manufacturer’s protocol. The run included 50% PhiX control library (Illumina #FC-110-3001). The output from the Illumina sequencing was processed as previously described,¹⁹ generating a table containing the sequences and the sequence count for all sequences in each sample.

Analysis pipeline

The data were analyzed using the in-house developed analysis pipeline previously described¹⁹ with minor modifications. The p values were adjusted using the Benjamini-Hochberg method with a false discovery rate of 0.05% and a 0.75 change in regression coefficient between COVID-19-positive and COVID-19-negative samples. The calculation of the Damerau-Levenshtein distance for the joining of highly similar sequences was done using the stringdist package.⁴⁰ The data were visualized in boxplots using the ggpubr package.⁴¹ Experimental scripts are available upon request.

Affinity purification of plasma proteins

Plasma proteins were affinity purified using 3′ biotinylated aptamers immobilized on streptavidin-coated magnetic beads incubated with pooled COVID-19-positive plasma. Aptamers (1 nmol RNA) were biotinylated in 1× reaction buffer (50 mM Tris-HCl [pH 7.5], 10 mM MgCl2, 10 mM DTT, 1.5 mM ATP, 50% PEG8000), 50 units T4 RNA ligase (Thermo Scientific, cat. no. EL0021), 6 μM final pCp-biotin (Jena Bioscience cat. no. NU-1706-BIO), and 0.5 units inorganic pyrophosphatase (Thermo Scientific cat. no. EF0221). The reaction was incubated for 16 h at 4°C. According to the manufacturer’s protocol, aptamers were purified using RNA Clean & Concentrator-25 spin columns (Zymo Research cat. no. R1017). Biotinylated RNA (400 pmol) was immobilized on 100 μL streptavidin magnetic beads (Invitrogen cat. no. 11205D) in 1× HBS buffer (150 mM NaCl, 20 mM HEPES [pH 7.4]) for 30 min. Beads were washed 3 times in 100 μL 1× HBS buffer supplemented with 1% tRNA, followed by incubation in 1× HBS buffer supplemented with 10% COVID-19-positive plasma for 30 min, and washed 3 times in 100 μL 1× HBS buffer. Bound proteins were eluted in 1× HBS buffer (20 mM EDTA) and incubated for 10 min at 45°C shaking.

MS

Sample preparation immunoprecipitations

20 μL of eluted sample from beads was mixed with 20 μL lysis buffer (6 M guanidinium hydrochloride, 10 mM tris(2-carboxyethyl)phosphine, 40 mM chloroacetamide, 50 mM HEPES [pH 8.5]). Samples were run through 10 kDa Spin filters (Amicon) at 14,000×g for 20 min. Filters were washed twice in 200 mL lysis buffer and centrifuged for 20 min at 14,000×g after each application. After washing, filters were turned upside down in a clean 1.5 mL Eppendorf tube, and samples were spun out for 1 min at 4,000×g. Samples were boiled at 95°C for 5 min and sonicated for 5 min in a water bath sonicator. 40 μL of the sample was diluted 1:3 with digestion buffer (10% acetonitrile in 50 mM HEPES [pH 8.5]), and proteins were digested with 100 ng LysC at 37°C for 3 h. Digestion buffer was added to a final dilution of 1:10 along with 50 ng trypsin, and samples were digested overnight at 37°C. Enzyme activity was quenched by adding 2% trifluoroacetic acid (TFA) to a final concentration of 1%. The peptides were desalted on a SOLAμ SPE plate (HRP, Thermo Scientific). The plate was spun down at 1,500 rpm for 1 min between each solvent application. The SPE filters were activated for each sample with 200 μL of 100% methanol (LC-MS grade, Sigma) and then 200 μL of 80% acetonitrile and 0.1% formic acid. The filters were subsequently washed twice with 200 μL of 1% TFA and 3% acetonitrile, after which the sample was loaded. After washing the filters twice with 200 μL of 0.1% formic acid, the peptides were eluted into 1.5 mL Eppendorf tubes using 40% acetonitrile and 0.1% formic acid. The eluted peptides were dried in a Thermo Scientific Speedvac and reconstituted in 12 μL 2% acetonitrile and 1% TFA containing iRT peptides for retention time normalization (Biognosys).

Immunoprecipitation MS data acquisition

From each sample, 300 ng of peptide mixture was analyzed by online nanoscale LC-MS/MS. Peptides were separated on a 15 cm C18-column (Thermo EasySpray ES804A) using an EASY-nLC 1200 system (Thermo Scientific). The column temperature was maintained at 30°C.

Buffer A consisted of 0.1% formic acid in water, and buffer B was 80% acetonitrile and 0.1% formic acid. The flow rate was kept at 250 nL/min, and the gradient started at 6% buffer B, going to 23% buffer B in 43 min. This was followed by a 12-min step going to 38% buffer B, followed by an increase to 60% buffer B in 5 min, and finally ramping up to 95% buffer B in 3 min and then holding it for 7 min to wash the column.

The Q Exactive Classic instrument (Thermo Scientific, Bremen, Germany) was run in data-dependent acquisition mode using a top-10 higher-energy collisional dissociation-MS/MS method with the following settings. The scan range was limited to 350–1,750 m/z. Full scan resolution was set to 70,000 m/z, with an automatic gain control (AGC) target of 3E6 and a maximum injection time (IT) value of 20 ms. Peptides were fragmented with a normalized collision energy of 25 eV, having a dynamic exclusion of 30 s (s), excluding unassigned ions and those with a charge state of +1. MS/MS resolution was set at 17,500 m/z, with an AGC target of 1E6 and a maximum IT of 60 ms.

Data analysis

All raw LC-MS/MS data files were processed together using Proteome Discoverer v.2.4 (Thermo Scientific) with the use of LFQ in both the processing and consensus steps. In the processing step, oxidation (M), protein N-terminal acetylation, and met-loss were set as dynamic modifications, with cysteine carbamidomethyl set as a static modification. All results were filtered with a percolator using a 1% false discovery rate, and the Minora Feature Detector was used for quantitation. SequestHT was used as the database, matching spectra against the human database from UniProt. The data were filtered for the common contaminants keratin, actin, and trypsin signals, and signals with an HT score <2 were excluded prior to data analysis. The percentage of total signal intensity was calculated for each identified protein and in each sample. The percentages were log2 transformed. The data were visualized in heatmaps by hierarchical cluster analysis using the pheatmap⁴² package.

Data and code availability

All data and R analysis scripts are available from the corresponding author upon reasonable request.