The upper level structure and design pattern of CIDO

Figure 1 lays out the high-level hierarchical structure of CIDO and shows the various imported external ontologies. Areas related to the coronavirus infectious disease represented by CIDO include: coronavirus taxonomy, coronavirus variants, genes and proteins and their mutations, phenotypes, diseases, epidemiology, diagnosis, host-coronavirus protein-protein interactions, vaccines, and drugs. All the terms are aligned under the top-level Basic Formal Ontology (BFO) (7) (Fig. 1). CIDO imports terms from over 20 reference ontologies from the OBO ontology library, with the representative ontologies introduced in Supplemental Table 2 and Fig. 1.

In addition to importing terms from existing ontologies, we have also generated many CIDO-specific terms e.g., resources for SARS-CoV-2 viral variants, amino acid mutations, and diagnostic medical device kits. New axioms, such as those linking different types of proteins and other molecules that are related to host-coronavirus protein-protein interactions (PPIs) and drug-target interactions, have also been developed for CIDO. In the version released on August 1, 2022, there are 370 relations used in CIDO, including 87 relations newly generated with “CIDO_” prefix. Admittedly, some of the newly generated relations in CIDO may be more suitable for the more general level Relation Ontology (RO) [27]; future research will involve further refinement of these relations.

Our previous Comment paper in Scientific Data [12] describes the general CIDO design pattern that lays out the relationships among selected major entities modeled in the ontology. In the next sections, we provide details of specific ontological modeling and representation provided in CIDO.

Ontological classification of coronaviruses and coronavirus variants

CIDO imports resources from the NCBITaxon to represent various coronaviruses and their relations [13]. SARS-CoV and SARS-CoV-2 belong to the Sarbecovirus, a subgenus of the genus Betacoronavirus. MERS-CoV belongs to Merbecovirus, a sibling to Sarbecovirus. Four human coronavirus strains (229E, NL63, HKU1, and OC43) cause mild common colds in humans, where 229E and NL63 belong to Alphacoronavirus, and HKU1 and OC43 belong to Embecovirus under Betacoronavirus.

We have generated 39 CIDO specific classes to represent specific COVID-19 viral variants. CIDO defines distinct viral variants of SARS-CoV-2 based on 3 classification methods: GISAID clades [35], PANGO lineages [36], and WHO clades [https://www.who.int/en/activities/tracking-SARS-CoV-2-variants/]. A viral variant is defined as a virus that has undergone variation such that there is a characteristic set of mutations in comparison to the reference virus sequence. These variants include various genetic mutations resulting in changes in transmission, infectivity, and virulence as compared to the original Wuhan reference strain. The GISAID clades and PANGO lineages both utilize the same data set but utilize different clustering algorithms to designate specific variants. PANGO lineages also differ by defining characteristic mutations that occur in a majority of specific SARS-Cov-2 variants while GISAID variants define universal mutations. The following examples illustrate these three hierarchies:

‘SARS-CoV-2 Delta virus’: ‘is a’ some ‘SARS-CoV-2 based on WHO classification’

‘SARS-CoV-2 BA.5 virus’ ‘is a’ some ‘SARS-CoV-2 based on PANGO lineage’

‘SARS-CoV-2 clade G virus’: ‘is a’ some ‘SARS-CoV-2 based on GISAID clades’

WHO utilizes GISAID clade and PANGO lineage representations as synonyms for epidemiologically relevant variants, designated either as a Variant of Concern (VoC) or as a Variant of Interest (VoI) [15]. VoIs are variants that are identified as having the potential to become VoCs through causing increased transmission or worse disease processes. VoCs remain designated as such until they are no longer prevalent.

Ontological representation of SARS-CoV-2 proteins and genes

CIDO imports terms for SARS-CoV-2 proteins from the Protein Ontology (PR) and terms for SARS-CoV-2 genes from the Ontology of Genes and Genomes (OGG), a simplified representation of which is shown in Fig. 2. Gene terms are based on those found in the NCBI Gene database [37] while proteins are as given by UniProtKB [38] [https://www.uniprot.org/uniprot/?query=proteome:up000464024], with cross-reference information from NCBI RefSeq [https://www.ncbi.nlm.nih.gov/protein?term=(sars-cov-2%20Wuhan-Hu-1%20AND%20refseq%5Bfilter%5D)]. CIDO represents only those genes that are described in NCBI Gene, and only those proteins (and their derivatives) that are described in UniProtKB. There are other protein open reading frames (ORFs) such as ORF2b (aka S.iORF1) [39], ORF-Sh and ORF-Mh [40], which are held in reserve, but they will be added should they gain experimental or database support. A full comparison between PR, RefSeq, and UniProtKB is given in Supplemental Table 3 with respect to accessions, genes, and names used (protein length and evidence for existence are also presented).

Fig. 2

SARS-CoV-2 proteins and genes. A PR modeling of SARS-CoV-2 proteins. B OGG modeling of SARS-CoV-2 genes. Black lines represent the ‘has gene template’ relation connecting proteins to genes. Red boxes denote proteins translated from ORFs that are internal to or overlap with those of the longer indicated gene (red arrows). The light blue box indicates proteins that are produced by proteolytic processing of either replicase polyprotein 1a or replicase polyprotein 1ab, while green boxes indicate those that derive specifically and uniquely from pp1a or pp1ab

In general, PR uses SARS-CoV-2 protein names as given in UniProtKB and gene names as given in RefSeq, wherever these are available. A key difference between the PR representation and those of RefSeq and UniProtKB is that the former has a single record for each proteolytic cleavage product of the ORF1ab (aka rep) gene, while each of the latter resources has two records for the subset of products that are encoded by both the polyprotein 1a (pp1a, aka ORF1a) and the polyprotein 1ab (pp1ab, aka ORF1ab) transcript (where the latter is the result of -1 ribosomal frameshifting). Both polyproteins are further processed by proteolytic cleavage; processing of either will yield ten identical chains (Fig. 2A, light blue box), while one additional chain is unique to ORF1a and five additional chains are unique to ORF1ab (green boxes). In addition, PR unites each of the polyproteins under the grouping term ‘rep gene translation product’ (the synonym is used here to prevent confusion with the ORF1ab transcript-derived polyprotein). Several proteins are translated from alternative ORFs within or overlapping transcripts that also produce longer proteins (red boxes). One of these, ORF9b, has been demonstrated (in SARS-CoV-1) to use leaky ribosome scanning [41]; potentially this mechanism applies to the others as well, though the existence of the ORFs labeled ‘putative’ is questionable [42]. All SARS-CoV-2 proteins are grouped under ‘severe acute respiratory syndrome coronavirus 2 protein’. In total—not counting the grouping terms—there are forty SARS-CoV-2-related PR terms. Currently, none of these represent proteoforms with amino acid modifications; these will be added in the future.

Ontological representation of SARS-CoV-2 amino acid variants

In addition to the representation of viral variants, CIDO also defines and represents various amino acid (AA) variants. Similar to the viral variant definition, an AA variant is defined in CIDO as “An amino acid in a protein that varies from another amino acid in comparison to the reference protein”. CIDO further defines the object property ‘is characteristic AA variant’ to describe a relation between an AA variant and a protein where the AA variant is a characteristic AA variant of a specific viral variant. An AA variant is defined as characteristic when the presence of the AA can be used to identify the AA variant. We characterize these variants by comparing the amino acid at a given position to the reference wild-type strain. For example, the D614G mutation in the spike polyprotein (S:D614G) is well known for emerging in several VoCs and has been proven to increase SARS-CoV-2 infectivity [43]. The CIDO class ‘D-614G in SARS-CoV-2 S protein’ (where S protein is just as the spike protein) has the following axioms (Fig. 2):

‘D-614G in SARS-CoV-2 S protein’:

‘characteristic AA variant of’ some ‘SARS-CoV-2 Omicron variant’

‘is a’ some ‘AA variant in SARS-CoV-2 S protein S1 RBD region’

‘has amino acid position’ value 614

‘has part’ some ‘glycine residue’

‘has mutated from’ some ‘aspartic acid’

However, the above framework does not work well for describing characteristic deletions or other mutation events. As the amino acid that was deleted does not exist, this leads to issues where the ontology asserts that something holds of ‘all coronaviral amino acids’. To address this issue, we define the AA deletion as a process. Moreover, this variation process can be generalized to include any mutation event. The relationship between the deletion process and a resulting AA variant, is defined as:

‘A888- deletion in SARS-CoV-2 S protein’: ‘is AA mutation of’ some ‘SARS-CoV-2 S protein’

as shown in Fig. 3.

Fig. 3

CIDO modeling of AA variants and mutations. CIDO represents AA variants as material entities if they are substitutions and AA mutations as processes to represent deletions in SARS-CoV-2 microbial variants. Both AA variants utilized analogous axioms due to differences in continuants and occurrents

Host phenotype modeling in CIDO

CIDO contains terms for 18 symptoms and 22 comorbidities commonly found in COVID-19 patients [44]. These symptoms and comorbidities are mapped to phenotypes in the Human Phenotype Ontology (HP) from where they are imported back into CIDO. To link these symptoms and comorbidities as they occur in relation to COVID-19, we have also generated new relations ‘disease susceptibly has phenotype’ and ‘disease susceptibly severe with comorbidity’. The first relation represents the relation between a disease process and a phenotype where the person with the disease is susceptible to having that phenotype. The second is a shortcut relation between a disease process which is susceptible to becoming more severe when the patient has the comorbidity. Examples of usage of these relations are:

CIDO also represents the relation between SARS-CoV-2 variant and specific phenotypes, for example, the relation between the Delta variant and the formation of syncytia in lungs [45]:

‘Delta variant disease process’: ‘bearer of disease susceptible to phenotype’ some syncytia

We are in the process of evaluating and submitting some of our newly generated relations to the OBO Relation Ontology (RO) as they may be more appropriate for inclusion there. For example, we have submitted two new relation terms ‘evolves into’ and ‘evolves from’ to the RO issue tracker (https://github.com/oborel/obo-relations/issues/620). If these relations are added to RO, we will then obsolete our original CIDO relation terms and replace them with the new RO terms.

Ontological modeling of epidemiology and public health

CIDO includes many terms related to the epidemiology of COVID-19, derived primarily from the Infectious Disease Ontology (IDO) [23] and the Virus Infectious Disease Ontology (VIDO) [14]. Recent research [46, 47] highlights the importance of viral load to SARS-CoV-2 transmission rates. Indeed, Wuhan, Delta, and Omicron strains are associated with distinct peak viral loads with respect to different demographics. VIDO characterizes ‘viral load’ as the proportion of virions to volume of a given portion of fluid in which the virions are located. VIDO provides a datatype property ‘has viral load measurement’ which supports representation of viral load values. For example, an instance of OBI’s class blood plasma specimen from an instance of a host infected by SARS-CoV-2 can be (partially) represented as having a viral load value in the following manner:

‘blood plasma specimen 1’ rdf:type ‘blood plasma specimen’

and ‘has part’ some ‘SARS-CoV-2’

and ‘has viral load measurement’ value 108

Additionally, VIDO provides virus-specific terminological content that can be extended in CIDO to represent other important epidemiological terms, such as COVID-19 prevalence, SARS-CoV-2 infectivity, and COVID-19 mortality rate.

Moreover, CIDO includes resources needed for comparison of transmission differences among SARS-CoV-2 variants. The Omicron variant is significantly more transmissible than the reference Wuhan strain and Delta strain. The transmission rate is often represented using R0, the basic reproduction number that measures the transmissibility of infectious agents [48]. The average R0 values for the Wuhan reference strain, Delta strain, and Omicron BA.1 strain are 2.69 [49], 5.02 [50], and 9.05 [51], respectively. Accordingly, we have generated a data property relation ‘has average R0’, which can be used to represent the R01 value of each variant:

‘SARS-CoV-2 reference strain: ‘has average R0’ value 2.69

‘SARS-CoV-2 Delta variant’: ‘has average R0’ value 5.02

‘SARS-CoV-2 Omicron BA.1 variant’: ‘has average R0’ value 9.05

COVID-19 diagnosis testing modeling in CIDO

During a pandemic, the availability of fast and accurate diagnostic testing is essential to control the situation. Because SARS-COV-2 is a novel virus, the traditional pathway to approve a testing kit to be used in the market will not satisfy the urgent demand in a timely manner. In the US, an Emergency Use Authorization (EUA) under Section 564 of the Federal Food, Drug, and Cosmetic Act (FD&C Act) allows the special authorization and use of drugs and other medical products during emerging infectious disease threats such as the COVID-19 pandemic. From 2020 March until now, the US Food and Drug Administration (FDA) has authorized hundreds of different types of in vitro diagnostic tests under the EUA authorizations. To make those EUA diagnostic testing data Findable, Accessible, Interoperable, and Reusable (FAIR) [28], it is important that the testing kits used are registered in a structured and machine-readable manner.

CIDO comprises representations of 345 molecular and serological diagnostic tests authorized by the FDA. We created a term ‘COVID-19 diagnostic testing device’ and its child term ‘FDA EUA authorized COVID-19 diagnostic testing device’, where the latter is to be the home of all FDA EUA authorized In Vitro Diagnostics (IVD) tests for COVID-19.

An example representation of the TaqPath COVID-19 Combo Kit from Thermo Fisher Scientific, Inc., which was authorized under an EUA authorization (https://www.fda.gov/media/136113/download) is shown in Fig. 4, which lays out the current CIDO representation of device, assay, diagnostic process and genes that the test is designed to detect. A device ‘TaqPath COVID-19 Combo Kit’ is ‘capable of’ a ‘COVID-19 RT-PCR assay’. This test detects the existence of N, S and ORF-1ab gene regions that are part of the corresponding genes of the SARS-CoV-2 reference strain. We created a short-cut relation ‘PCR kit detects gene’ to represent a direct relationship between a diagnostic testing kit and the target gene/sequence fragments. Another short-cut relation ‘device utilizes material’ was created to link the diagnostic testing and the tested specimen. This relation can be logically represented as a property chain (https://github.com/oborel/obo-relations/issues/497):

Fig. 4

Modeling of COVID-19 diagnostic testing using CIDO. *, only two out of six specimen terms are shown in this figure

This particular diagnostic testing kit can utilize 6 specimen types, as again shown in Fig. 4. The following axiom represent the ontological arrangement of such a relation using a union of 6 specimen terms:

‘device utilizes material’ some (‘nasopharyngeal swab specimen’ or ‘oropharyngeal swab specimen’ or ‘anterior nasal swab specimen’ or ‘mid-turbinate nasal swab specimen’ or ‘nasopharyngeal aspirate specimen’ or ‘bronchial alveolar lavage’)

Using the strategy defined here, we systematically collected and used CIDO to model and represent over 300 molecular and serological diagnostic tests, including 225 SARS-CoV-2 RT-PCR assays, authorized by US FDA. All the 343 tests are annotated with a total of ten COVID-19 diagnostic technologies, such as RT-PCR, LAMP, Next Generation Sequencing, a CRISP-based method, ELISA, lateral flow immunoassay, chemiluminescent, and so on.

CIDO modeling and representation of host-coronavirus protein-protein interactions and drugs

CIDO represents over 300 experimentally verified host-coronavirus protein-protein interactions (PPIs), over 300 anti-coronaviral chemicals and/or their corresponding drugs, and over 400 drug targets. Here the coronaviral proteins may derive from SARS-CoV, MERS-CoV, or SARS-CoV-2. In early 2020, we performed literature mining and identified 110 chemical drugs and 26 antibodies effective, either in vitro or in vivo, against at least one human coronavirus infection, where the human coronaviruses involved are primarily SARS-CoV and MERS-CoV [52]. Our ontological representation, classification, and analysis of these drugs yielded many potentially valuable scientific insights. Since early 2020, we have collected more drugs and chemicals with a focus on those against SARS-CoV-2. Furthermore, we have collected and annotated representations of further PPIs and chemical-drug interactions.

All CIDO-represented host-coronavirus PPIs are experimentally verified and reported in the literature. For example, CIDO has recorded 332 physically associated PPIs identified by the affinity-purification mass spectrometry assay [5]. These PPIs involve both proteins from the SARS-CoV-2 side and the host side, and many of these coronaviral and host proteins are also targets of multiple drugs.

In CIDO, each host-coronavirus PPI is defined to have at least two participants, including one protein from a coronavirus and one from its host. For example, the ‘host-SARS-CoV-2 protein-protein interaction’ is defined as:

(‘has participant’ some ‘SARS-CoV-2 protein’) and (‘has participant’ some (organism and ‘has role’ some ‘host role’))

Figure 5 illustrates how CIDO represents hundreds of host-SARS-CoV-2 PPIs, drug active ingredients, and chemical-protein interactions. Specifically, there are three specific PPIs under the class ‘SARS-CoV-2 nsp5 protein interaction with host protein’, such as ‘SARS-CoV-2 nsp5 protein binding to human HDAC2’. This example PPI has two participants:

‘has participant’ some ‘3C-like proteinase (SARS-CoV-2)’

‘has participant’ some ‘histone deacetylase 2 (human)’

Note that 3C-like proteinase, another name for nsp5, can be inhibited by the chemical nirmatrelvir, a component of the Pfizer drug Paxlovid. Human histone deacetylase 2 (i.e., HDAC2), can be inhibited by a chemical ‘Valproic Acid’, which has been found valuable against SARS-CoV-2 [53]. These relations are logically defined in CIDO as follows (Fig. 5B and C):

'nirmatrelvir': ‘chemical inhibits protein’ some ‘3C-like proteinase (SARS-CoV-2)’

‘Valproic Acid’: ‘chemical inhibits protein’ some ‘histone deacetylase 2 (human)’

Fig. 5

Host-coronavirus protein-protein interactions (PPIs) and drugs targeting the viral or host proteins. A The hierarchy of PPIs, including ‘SARS-CoV-2 nsp5 protein binding to human HDAC2’. B The chemical nirmatrelvir (a component of the Pfizer drug Paxlovid) is an inhibitor of the virus protein nsp5 (i.e., 3C-like proteinase), which is critical for viral replication. C A chemical ‘Valproic Acid’ is an inhibitor of the HDAC2 (i.e., histone deacetylase 2). Valproic acid is also a valuable candidate against SARS-CoV-2

Anti-coronavirus vaccine representation in CIDO

As the developers of the Vaccine Ontology (VO) [54], we (YH, AL, AH, PH) first represented a total of over 100 COVID-19 vaccines at different stages (licensed, authorized, in clinical trials, or verified with laboratory animal models) in VO, and then imported these terms from VO to CIDO (Fig. 1, Supplemental Table 2). In total, we have imported over 300 terms from the VO to CIDO. Furthermore, we have developed Cov19VaxKB, a web-based Integrative COVID-19 vaccine knowledge base, which has used ontologies including the VO to represent, classify, and analyze various COVID-19 vaccines and vaccine components (e.g., vaccine adjuvants), and vaccine adverse events [55]. We have also developed reverse vaccinology and machine learning methods to predict vaccine antigen candidates [56]. The functions and immune mechanisms of these candidates are being further analyzed using ontology-based approaches [15]. Furthermore, we have been using CIDO and other ontologies including the Ontology of Adverse Events (OAE) to systematically examine adverse events associated with SARS/MERS/COVID-19 vaccine candidates.

Clinical metadata type representation in CIDO

To support classification and analysis of clinical data, CIDO includes representations of many clinical metadata types. Metadata is the data that provides information about other data. In our study of COVID-19 related clinical data, we have focused on two use cases: the analysis of vaccine adverse events using the VAERS data resource as described above and the analysis of the clinical data from the National COVID Cohort Collaborative (N3C) program [57]. The N3C system is a collection of harmonized clinical data on COVID-19 from contributing data partners. N3C data is represented using the OMOP common data model (CDM). From the OBO ontology point of view, OMOP has its issues such as the lack of semantics, ambiguities, and hidden assumptions [58]. In our N3C related clinical data study, we have focused on the mapping of the OMOP CDM elements and OBO ontologies and adding semantic relations among terms.

Table 1 lists the representative clinical metadata types that are primarily mapped to the OMOP CDM elements. These are general clinical data types applicable to studies not only of COVID-19 but also of other human diseases. As a result, all these terms are imported from other reference OBO ontologies. The Ontology of Precision Medicine and Investigation (OPMI) [59, 60], another OBO library ontology, has been used as a major reference ontology to represent those clinical data types not found in other OBO ontologies (Table 1). After the mapping of OMOP CDM elements to OBO ontologies, we imported these mapped terms to CIDO to support COVID-19 clinical data annotation and analysis.

Table 1 Representative clinical metadata types covered in CIDO. All listed examples are considered classes in the ontology

In the OMOP / N3C data structure, each concept set groups terms into what are called value sets. A value set is a set of codes selected from those defined by one or more code systems to specify which codes can be used in a particular context. However, their grouping is heuristic and not ontology-based. The ontology support is an ongoing project. OMOP2OBO is the first health system-wide integration and alignment system that systematically maps over 23,000 concepts from OMOP standard clinical terminologies to OBO concepts [61]. While OMOP2OBO is more focused on the value set mapping, our mapping and further term generation (Table 1) is more focused on the small set of the core OMOP CDM concept set meta elements. The two complementary systems can be used together to support robust clinical COVID-19 data annotation, integration, and analysis.

Visual evolution analysis of CIDO

To provide a condensed and comprehensive visualization of CIDO, we have previously developed a new Weighted Aggregate Partial-Area Taxonomy (WAT) summarization network method and used it to analyze an early version (version 1.0.108) of CIDO with a total of 5138 concepts [34]. Since then, newer versions of CIDO that include more concepts have been generated. To evaluate these new additions to CIDO, we have generated a new WAT summarization network that visualizes CIDO version 1.0.306 with 10,853 concepts (Fig. 6). As shown in Fig. 6, major branches of CIDO include infectious diseases, genes, vaccines, chemicals, and COVID-19 testing devices.

Fig. 6

The weighted aggregate taxonomy (WAT) for CIDO (version 1.0.306) with 10,853 concepts (b = 42). A white node inside a colored rectangular box represents a partial-area, which is a group of concepts having the same set of nonhierarchical (lateral) relationships and similar semantics denoted by the concept listed inside the white node. Relationships are listed inside the colored box (inherited ones are not shown). The boxes are color-coded by cardinalities of their sets of lateral relationships. Upward arrows are the hierarchical relationships connecting partial-areas. The weight of a partial-area is defined as the number of descendant concepts. A partial-area with a weight less than b is small and is aggregated into its closest ancestor large partial-area. A large partial-area having no aggregated partial-areas is represented as a rectangle white box with one number indicating the number of summarized concepts. A large partial-area having aggregated partial-areas is represented as a rectangle with rounded corners and with three numbers. The first number inside () is the number of summarized concepts including concepts aggregated from small partial-areas, the second number inside is the number of small partial-areas aggregated into it, and the third number inside [] is the number of concepts of the partial-area before the aggregation. See more details in Supplemental File 1

Comparing the old version (Fig. 2 in Supplemental File 1) with the new, we can identify which nodes had a considerable increase in the number of new descendant terms. For example, “COVID-19 vaccine” (120) [72] has been added to the ontology visualization (Fig. 6). The number (120) means that the term “COVID-19 vaccine” includes 120 descendant terms, with 48 of those aggregated from 48 descendant nodes of “COVID-19 vaccine,” each of which has only one term (less than b = 42), and 72 representing all other descendant terms of the large partial-area “COVID-19 vaccine” before the aggregation. By expanding this node in the manner supported by the OAF tool, we can see some interesting newly added vaccine terms such as “Pfizer–BioNTech COVID-19 vaccine”, “Moderna COVID-19 vaccine”, “Oxford–AstraZeneca COVID-19 vaccine”, and “Nanocovax”. In contrast, the old version includes only one term for “COVID-19 vaccine” without any descendant term. Another example is “FDA EUA authorized COVID-19 diagnostic testing device” (345)[116] in Fig. 6 including terms “COVID-19 Nucleic Acid RT-PCR Test Kit” and “BinaxNOWTM COVID-19 Ag Card Home Test” for which there are no corresponding terms in the old version.

Use cases of CIDO

CIDO has been proposed and used in many applications by us or the wider community as exemplified by references [15, 44, 52, 62,63,64,65,66,67]. Five use cases of our own application of CIDO are introduced here.

(1)

Ontology-based coronavirus-related knowledge and data standardization, annotation, mapping, integration, and inferencing, supporting advanced COVID-19 data analysis

As a reference ontology in the field of coronavirus infectious disease, CIDO provides a standard representation and definitions of terms and axioms in various areas related to COVID-19 and other coronavirus diseases. The above sections have provided details on how CIDO standardizes and classifies terms and relations in different domains related to coronavirus diseases. Usage of the CIDO standard representation enhances data FAIRness, annotation, and integration.

The COVoc Controlled Vocabulary for COVID-19 is an application ontology developed by the European Bioinformatics Institute (EMBL-EBI) and the Swiss Institute of Bioinformatics (SIB) in March 2020 [