Investigating in-hospital transmission of Clostridioides difficile from asymptomatic patients

Abstract

Background:

Clostridioides difficile is a bacillus that can colonize the intestinal tract. C. difficile infection (CDI) is a common, health care–associated disease often coinciding with antibiotic use, with presentation ranging from diarrhea to toxic megacolon and death. However, individuals can carry the pathogen without exhibiting symptoms of disease. C. difficile carriers may serve as an important reservoir for in-hospital transmission of C. difficile. The objective of this study was to assess the role asymptomatic C. difficile carriers may play in pathogen transmission.

Methods:

In this retrospective cohort study, rectal swabs collected to test for antimicrobial-resistant organisms either upon admission or during the hospital stay were used to test for asymptomatic C. difficile carriage using molecular testing, with positive samples cultured and subsequently sequenced. Sampling occurred across three tertiary care hospitals with predominantly multi-bedded rooms. A whole-genome single-nucleotide variant phylogenetic tree was then constructed and closely related samples from patients that had spatial and temporal in-hospital overlap were identified as putative transmission events.

Results:

Approximately 11% of the 1,467 patients tested were carriers of C. difficile. Seventy-four of the carriers were culture-positive, of which 43 had a suitable sequence for typing. Of the sequenced samples, 40% were identified as belonging to a potential transmission event. Six different potential transmission groups were identified, with the largest putative transmission group spanning two hospitals and 6 months.

Conclusions:

Our results suggest that asymptomatic transmission in hospital settings may be much more common than previously thought, with asymptomatic individuals colonized with C. difficile providing a sizable source of transmission of the pathogen.

Introduction

Clostridioides difficile is a gram-positive, anaerobic, and spore-forming bacillus that can either colonize or infect patients (1). C. difficile infection (CDI) is the most common cause of hospital-acquired diarrhea and can result in toxic megacolon and death (2,3). CDI causes a significant financial burden to the economy, with a 2017 study by Wilcox et al estimating the cost in the United Kingdom to be £6,294 per case of CDI and £7,539 per case of recurrent CDI (4). C. difficile is transmitted by the fecal-oral route, mainly in the form of spores that traverse the stomach and germinate in the small intestine (5). Antibiotic use results in dysbiosis of the gut microbiota, allowing for C. difficile to more easily colonize and infect the colon (1,5), and is as such considered the most important risk factor for CDI (6). Although mainly considered a health care–associated disease, CDI cases have been seen at the community level in individuals lacking traditional risk factors (7).

C. difficile strains are often classified by ribotype, with ribotype 027 being considered one of the more virulent strains due to their higher production of toxins (8). Ribotype 027 outbreaks have typically resulted in worse health outcomes including higher death rates than other strains (9). This strain was first reported in the United Kingdom in 2005 (10), before rapidly spreading to other European countries (11). While recent North American studies have found that the prevalence of ribotype 027 across North America has decreased significantly in recent years, it remains the most common strain type in C. difficile epidemics (9,12). The prevalence dramatically increased in Europe from 2008 to 2012–2013, although it has showed a decline in prevalence from 2013 onwards (11,13).

At present, testing and treatment for C. difficile is only recommended for those presenting with symptoms of a potential CDI (14). Colonized individuals outnumber those who are symptomatic (ie, infected) (15,16), however, and the strains in these individuals are genetically similar to those found in symptomatic individuals (17). Asymptomatic individuals who are carriers of C. difficile can shed the pathogen and colonization rates of toxigenic C. difficile strains among asymptomatic carriers have been reported to be greater than 10% (16,17). It has been proposed that identifying asymptomatic carriers of C. difficile may reduce transmission in health care settings (18–20). However, it is unclear how frequently transmission events from asymptomatic patients occur. One previous study found that carriage of C. difficile was a significant risk factor for CDI in both the carriers themselves and noncarriers who share the same physical space as the carrier; however, this study did not carry out whole-genome sequencing to identify exact transmission patterns (21).

The main goal of this study was to determine whether asymptomatic patients harbouring C. difficile may contribute meaningfully to in-hospital transmission of C. difficile using whole-genome sequencing, as evidenced by genetic relatedness and spatiotemporal clustering.

Methods

This retrospective cohort study was conducted in three tertiary care hospitals within one city in southwestern Ontario, Canada with predominantly multi-bedded rooms. As per policy, patients with antimicrobial-resistant organisms such as MRSA and vancomycin-resistant Enterococcus (VRE), or with CDI, were cared for with contact precautions, ie, using gloves and gowns. Clinical data on patients eligible for this study were retrospectively collected from medical charts.

During the study period from September 2018 to May 2019, swabs routinely collected and tested for methicillin-resistant Staphylococcus aureus and vancomycin-resistant Enterococci screening upon admission or for point prevalence testing for MRSA or VRE outbreak investigations and management were then additionally tested for C. difficile on selected hospital wards with historically the highest CDI rates. At that time, all patients meeting criteria for increased risk of MRSA/VRE colonization (such as having a previous hospital admission within the previous 2 years or with a history of antimicrobial-resistant pathogens) were routinely swabbed upon admission. Patients with symptomatic CDI at the time of sampling were not included in the study. Swabs were stored in 2 mL of E-swab transport medium (Copan Diagnostics, Murrieta, CA, USA) at 4°C for up to 3 months until testing. DNA was extracted by vortexing the sample then diluting 100 µL of E-swab media 1:1 with 100 µL Chelex Lysis Solution (Sigma-Aldrich, St. Louis, MO, USA). The Chelex preparations were then vortexed before being heated at 95°C for 15 min. After heating, samples were vortexed again before centrifugation for 2 min at 13,000 × g. Loop-mediated isothermal amplification (LAMP) was then performed using a laboratory-developed assay (22) targeting a 240 bp fragment of the TcdC gene using 5 µL of crude lysis supernatant as template. LAMP was performed using a RotorGene Q (Qiagen, Germantown, MD, USA). The majority of LAMP-positive samples and an equal number of samples negative for TcdC via LAMP were further tested by a laboratory-developed multiplex real-time PCR assay targeting the TcdA and TcdC genes (23) to capture samples that may have been TcdC-negative in LAMP. Extracted DNA was amplified through detection of a 226 bp fragment of the TcdC gene, a 139 bp fragment of the TcdA gene, and a 94 bp segment of a eubacterial ribosomal gene (RrnB) (23). The real-time PCR assay was also performed using a RotorGene Q.

All samples that were both LAMP and PCR positive were cultured on C. difficile selective CHROMagar (Micronostyx, Ottawa, ON, Canada). In the interests of cost efficiency and to not waste resources, we only cultured specimens that we believed would produce useful results. By using first LAMP and then PCR, we were able to have two levels of confirmation that we were likely to successfully culture a sample. The goal of culturing samples was not as a test of whether a patient was colonized with C. difficile; but rather to grow the bacteria from which we could then extract DNA to perform genomic analysis. We vortexed 30 µL of E-swab media for 10 s and then inoculated onto plates and incubated for 24 h in a Bactron IV anaerobic chamber (Sheldon Manufacturing, Cornelius, OR, USA). Colonies were visualized under a fluorescent light and a proportion were verified by MALDI-TOF (bioMérieux VITEK MS, Marcy-l'Etoile, France). For sequencing, isolated subcultures were serially subcultured three times using blood agar media.

Once cultured, genomic DNA extraction was performed, before samples were prepared for sequencing as previously described (24). Briefly, genomic DNA was fragmented and adaptors were ligated onto the fragmented DNA. Samples were then treated with uracil-specific excision reagent before sample purification and size selection. Barcodes were then attached to samples before a final sample purification and dual size selection. Sequencing was then carried out using 2 × 150 bp paired end sequencing on an Illumina NextSeq (Illumina, San Diego, CA, USA) sequencer. After sequencing, FASTQ files were analyzed via FastQC (25), barcode and adaptor sequences were removed using Trimmomatic (26), and SPAdes (27) was used to assemble genomes, before Kraken2 (28) was used to confirm that all sequences were C. difficile genomes. Minimap2 (29) and subsequently samtools (30) were used to assess depth and breadth of coverage. Parsnp (31) was then used to align genomes and extract single-nucleotide variants (SNVs) uninvolved in lateral gene transfer, before construction of maximum-likelihood phylogenetic trees using the RAxML-HPC BlackBox platform with the GTRGAMMA + I substitution model and automatic bootstrapping (32,33). Further genomic analysis was performed to identify pairwise core genome SNV distances between samples using snippy (https://github.com/tseemann/snippy) and then snp-dists (https://github.com/tseemann/snp-dists). A potential direct transmission event was predicted when there were ≤2 core genome SNV differences (34) between samples and there was an epidemiological link between the samples.

Using the genomic data, multilocus sequence typing was additionally determined using the pubMLST database using the schema developed by Griffiths et al (35).

Results

A total of 1,467 patients were tested for C. difficile colonization using rectal swabs. In total, 50 of the 1,467 patients (3.4%) were diagnosed with symptomatic CDI at some point during their hospitalization, with 4 of the tested patients being diagnosed with symptomatic CDI upon admission.

Of the 1,467 tested patients, 167 (11.4%) tested positive by LAMP for C. difficile colonization, with 12/167 (7.2%) of these carriers being diagnosed with symptomatic CDI at a later point during their hospitalization. These 167 patients that tested positive do not include the four that were symptomatic upon admission. Among a subset of 231 specimens (the 167 LAMP-positive samples and an additional 64 LAMP-negative samples) tested by both PCR and LAMP, 79 were concordant positives and cultured on CHROMagar media. Demographic data were available for 77 of the samples tested by both LAMP and PCR that were identified as negative. Of these samples, 42% were female with an overall average age of 71.1 (SD 15.5) years. Seventy-four of the 79 concordant LAMP- and PCR-positives were successfully cultured (93.7%; 42% female, average age 70.5 (SD 15.5) years). Of the 74 culture-positive samples, 43 (58.1%) were successfully sequenced with >88% coverage of the reference genome as determined by minimap2 and samtools and identified as C. difficile via Kraken2. Average coverage was 94.8% (SD 2%) and average depth of coverage was 137.1-fold (SD 103.6-fold) (Supplementary Table 1). Only one (2.3%) of these 43 successfully sequenced samples was associated with a symptomatic CDI case who became symptomatic after the test for colonization occurred. The majority of successfully sequenced samples were obtained 2 or more days after admission (28/44, 63.6%). No outbreaks of CDI were observed in the study hospitals during the study period.

When a maximum-likelihood whole-genome SNV phylogenetic tree was constructed, samples clustered together monophyletically by multilocus sequence type (MLST) (Figure 1). Analysis of patient genomic data (core genome SNV analysis using snippy), combined with spatial and temporal metadata of patient hospitalizations, revealed six clusters suggestive of transmission (Table 1). Pairwise core genome SNV distance matrices are included in Supplementary Tables 2–11, separated by MLST. Unrelated samples within an MLST differed by an average of 36.8 (SD 35.4) SNVs. Assuming that each putative transmission group was independent and C. difficile was transmitted from one member of the transmission group to another member(s), 40% of samples were linked to a potential in-hospital transmission event, including the index cases, which were assigned as the first patient in a transmission group to test positive. However, there were several additional samples that clustered together on the tree, suggesting potential community transmission or transmission in another congregate setting, along with unidentified intermediate carriers or environmental sources facilitating transmission within the health care facility. Three samples, all identified as MLST 1, clustered together in the tree, but there was no spatial or temporal overlap in the associated hospital stays of the patients from which these samples originated and additionally differed by greater than two SNVs in the core genome. The only sequenced genome associated with symptomatic CDI was MLST type 2 and was a member of transmission group 1. There were 17 different MLSTs found among our samples (Table 2).

Figure 1: Maximum-likelihood genome-wide single-nucleotide variants phylogenetic tree of sequenced samples.

Note: Colour boxes containing groups of samples indicate the putative transmission groups, with numbers within circles indicating the transmission group identifier (detailed in Table 1); bootstrap values are shown at each node; multilocus sequence type numbers are indicated on the right side of the figure but were not used in assessing transmission

Table 1.

Putative C. difficile transmission events

Transmission group (non-ordinal)	Sample ID	Hospital ID	MLST	Admission	Discharge	Positive test	Dates of hospitalization overlap
1	7	1	2	Sept 23, 2018	Oct 16, 2018	Oct 10, 2018	Sept 29 to Oct 16 with sample 85
	85			Sept 29, 2018	Nov 25, 2018	Nov 13, 2018	Sept 29 to Oct 16 with sample 7; Nov 7 to Nov 21 with Sample 27
	27			Nov 7, 2018	Nov 21, 2018	Nov 13, 2018	Nov 7 to Nov 21 with Sample 85
	81	2		Mar 10, 2019	May 8, 2019	Mar 25, 2019	Mar 12 to Apr 24 with sample 13
	13			Mar 12, 2019	Apr 24, 2019	Mar 12, 2019	Mar 12 to Apr 24 with sample 81
2	146	1	110	Oct 5, 2018	Nov 9, 2018	Oct 15, 2018	Oct 30 to Nov 9 with sample 4
	4			Oct 30, 2018	Nov 19, 2018	Nov 5, 2018	Oct 30 to Nov 9 with sample 146
	98			Dec 5, 2018	Dec 18, 2018	Dec 10, 2018	Only spatial overlap with sample 4; hospitalization differed by 17 days
3	44	1	42	Oct 24, 2018	Jan 10, 2019	Nov 6, 2018	Dec 14 to Jan 10 with sample 61
	61			Dec 14, 2018	Jan 17, 2019	Dec 15, 2018	Dec 14 to Jan 10 with sample 44
4	99	1	8	Oct 26, 2018	Nov 7, 2018	Oct 30, 2018	Only spatial overlap with sample 28; hospitalization differed by 70 days
	28			Jan 16, 2019	Jan 27, 2019	Jan 17, 2019	Only spatial overlap with samples 28 and 86
	86			Feb 6, 2019	Feb 21, 2019	Feb 7, 2019	Only spatial overlap with sample 28; hospitalization differed by 10 days
5	151	1	58	Feb 9, 2019	Mar 19, 2019	Mar 11, 2019	Feb 21 to Mar 19 with sample 153
	153			Feb 21, 2019	Mar 19, 2019	Mar 3, 2019	Feb 21 to Mar 19 with sample 151
6	17	2	12	Jan 17, 2019	Jul 9, 2019	Mar 18, 2019	Jan 17 to Jul 9 with sample 39
	39			Jan 17, 2019	Jul 9, 2019	Feb 27, 2019	Jan 17 to Jul 9 with sample 17

Transmission groups are not ordered chronologically, but within a transmission group, samples are ordered chronologically by admission date. A transmission group is listed here if samples differ by ≤2 core genome SNVs and there was spatial and temporal overlap between the members of a transmission group (ie, patients were in the same ward in the same hospital at the same time)

MLST = Multilocus sequence type

Table 2.

MLST counts and prevalence rates for sequenced C. difficile samples

MLST	No (%); N = 43
1	3 (7.0%)
2	6 (13.9%)
3	1 (2.3%)
7	2 (4.6%)
8	3 (7.0%)
10	2 (4.6%)
12	3 (7.0%)
14	2 (4.6%)
18	1 (2.3%)
42	5 (11.6%)
43	1 (2.3%)
46	1 (2.3%)
58	4 (9.3%)
110	5 (11.6%)
123	1 (2.3%)
192	1 (2.3%)
302	2 (4.6%)

MLST = Multilocus sequence type

Patients within a single assumed transmission group had at least one instance of temporal overlap in the same ward as another patient from the same transmission group, with the exception of two patients, who did not have in-hospital spatial overlap with any of the three patients within the same transmission group based on sequencing alone. This is likely suggestive of community transmission, as on the genomic level these samples are all closely related.

All genomic data analyzed in this work are available in NCBI BioProject PRJNA1087383.

Discussion

The results of the present study indicate that asymptomatic patients may be a significant reservoir of C. difficile in hospitals, with just under half of all carriers in our study being associated with a potential patient-to-patient transmission event. We were able to identify these potential patient-to-patient transmission events through the combination of whole-genome sequencing, spatial, and temporal data. We have also found that the current regional genetic landscape of C. difficile is very diverse, with a variety of MLSTs identified in our samples.

Two transmission groups were more complex in terms of epidemiological links than others. First, transmission group 1 spanned two hospitals and at least 6 months. Transmission from patients in hospital 1 (positive tests in October and November 2018) to patients in hospital 2 (positive tests in March 2019) likely occurred in a secondary community setting (such as a long-term care home) as the samples were very closely related with ≤3 pairwise SNV differences rather than ≤2. Based on hospitalization dates and pairwise SNV differences, we believe that the patients from hospital 1 were first colonized before transmission occurred in the community to one of the patients hospitalized at hospital 2 months later, who then transmitted the pathogen to at least one more patient. Given the long timeframe, we modified the threshold of the pairwise difference when defining this transmission group given the epidemiologic link. Importantly, it is possible that there was not direct transmission between the two subgroups of transmission group 1; rather there may have been intermediate carriers that were not hospitalized and thus not sampled in the present study, or shared environmental sources.

Second, transmission group 4 also contained a sample not obviously epidemiologically linked to the other two samples within the group. However, deeper analysis revealed that all three samples had spatial overlap in hospital 1, despite the long timeframe between positive samples (October 30, 2018; January 17, 2019; February 7, 2019). This could be explained by instances of transmission occurring during previous hospitalizations. The number of pairwise SNV differences combined with the hospitalization dates was suggestive of transmission from patient 99 to 28, and then from patient 28 to 86, which may also explain why the difference in SNV was 3 rather than 2 between the presumed index (patient 99) and the last patient in the sequence (patient 86). Again, it is important to consider that there may also have been unsampled intermediate carriers or environmental sources within this transmission pathway.

A Canada-wide study from the same time period as the present study found that the top three most common ribotypes of C. difficile in pediatric patients were ribotype 106, 020, and 014, with ribotype 027 significantly less prevalent (36). This aligns with our observed result of only identifying 7% of samples as MLST1, which is the MLST that correlates to ribotype 027. This is contrasted with an earlier Canadian study concluding in 2019 that found ribotype 027 comprising 16% of CDI cases (37). However, it is important to note that this latter rate was found in symptomatic rather than asymptomatic patients. Interestingly, we did not identify any potential transmission events with ribotype 027, but instead we identified potential transmission events of ribotypes not necessarily reported to be associated with significant transmission in health care settings.

Whole-genome sequencing can be used to evaluate potential transmission in a wide variety of pathogens, and as such has been used in several previous studies to identify instances of in-hospital patient-to-patient transmission, such as Simon et al, who were able to identify health care-associated clusters of Staphylococcus epidermidis (38). Furthermore, one recent study used whole-genome sequencing to identify in-hospital transmission events of antimicrobial-resistant Klebsiella pneumoniae (39). The present study is also a successful example of using whole-genome sequencing to estimate the likelihood of patient-to-patient pathogen transmission.

In the United Kingdom, Eyre et al found that 11% of patients were asymptomatically colonized with C. difficile, of which 72% harboured toxigenic strains, although no clear evidence of transmission from an asymptomatic individual to another individual was found despite serial samples being taken, which may have been due to the small sample size of the study (132 patients provided samples) (40). These carriage and toxin rates were similar to a study from the United States that identified 7.4% of ICU patients as asymptomatic carriers, with 69% of sequenced isolates carrying a toxin gene, supporting the conclusion that asymptomatic carriers are an important reservoir of toxigenic C. difficile in hospitals (41). A further study from the United States found that 51% of asymptomatic patients were carriers of toxigenic C. difficile, and that of these, 37% harboured epidemic strains (ribotype 027). In addition, 7% of patients had symptomatic CDI and asymptomatic carriers outnumbered patients with symptomatic CDI 7 to 1, with the conclusion that asymptomatic carriers of both epidemic and nonepidemic C. difficile strains contribute significantly to disease transmission (18). This reported rate of colonization was much higher than our observed rate of 11%, although these higher rates were from long-term care home residents housed in two wards of the same building and screening used stool samples combined with rectal swabs, which may explain their higher reported rate of colonization. Our study used rectal swabs, which have previously been demonstrated to detect 80% of gut pathogens identified from stool samples (42).

A previous study found that asymptomatic carriers of C. difficile contribute meaningfully to in-hospital transmission, finding that 25% of cases of symptomatic CDI were epidemiologically linked to a known asymptomatic patient; however, whole-genome sequencing was not conducted (43). This percentage is comparable to our observed rate of 26% of colonized patients being linked to known asymptomatic patients, found when the arbitrarily assigned index cases are removed from transmission groups. Curry et al also noted that transmission still occurred even when carriers were not physically in the same environment at the same time as the patient to whom they transmitted C. difficile, suggesting that asymptomatic carriers are able to release C. difficile into the hospital environment where the bacteria can remain before colonizing a new carrier (43). This is likely to have occurred in our present study as well, as some genetically and spatially linked cases did not have significant temporal overlap in the hospital.

Similarly, a study in a long-term care facility found that residents with asymptomatic carriage of C. difficile were a substantial source of transmission to other residents based on whole-genome sequencing (44). Durham et al determined that individuals with symptomatic CDI transmit C. difficile at a rate 15 times greater than asymptomatic carriers (45). Furthermore, they found that long-term care facility residents transmit C. difficile at a rate that is 27% that of hospitalized patients, and that community transmission is 0.1% that of in-hospital transmission. However, due to the larger reservoir of possible community and asymptomatic transmission, these still represents a significant source of C. difficile transmission (45). Durham et al also suggested that control strategies to reduce community transmission have the potential to reduce hospital-onset CDI by reducing the number of asymptomatically colonized individuals entering the hospital (45).

Our study has several limitations. First, given the study design, we were unable to have a consistent pattern of multiple tests per patient, eg, upon admission, during hospital stay, and upon discharge; the data presented here only represents one swab per patient. Hence, there remains uncertainty around the timing and place of acquisition. It is also important to note that transmission may have not occurred directly from patient to patient; environmental reservoirs of C. difficile may have played a role in facilitating transmission (46). However, whole-genome sequencing does suggest transmission of some nature within our studied hospitals. Furthermore, as rectal swabs were used rather than stool samples, this may have led to an underestimation of the true incidence and therefore transmission rates of C. difficile. Not all PCR-positive samples were successfully cultured, and not all culture-positive samples were successfully sequenced, which led to a reduction in our dataset that may have resulted in further underestimations of transmission events. Additionally, stool samples were not collected for sequencing from CDI patients concurrently with asymptomatic patients, so no genomic comparison was made between the strains present in asymptomatic carriers and symptomatic CDI patients.

Conclusions

While C. difficile colonization in admitted patients is much more prevalent than patient-to-patient in-hospital transmission, our study suggest that potential transmission among asymptomatic individuals might occur more frequently than previously thought, as evidenced by the spatiotemporal clustering of genetically related C. difficile specimens. Further studies are required to determine whether routine detection of colonization could reduce overall rates of CDI.

Funding Statement

This study was supported by the Hamilton Health Sciences Research Strategic Initiative Program (grant to DM). SJCB was funded by a MITACS Elevate Fellowship. AGM was supported by a David Braley Chair in Computational Biology.

Contributors:

Conceptualization, SJC Baker, J Maciejewski, J Gilchrist, M Smieja, MB Loeb, AG McArthur, D Mertz; Methodology, SJC Baker, J Maciejewski, MT Usuanlele, J Gilchrist, M Smieja, MB Loeb, AG McArthur, D Mertz; Software, AG McArthur; Validation, AG McArthur; Formal Analysis, SJC Baker, MG Surette, AG McArthur, D Mertz; Investigation, SJC Baker, J Maciejewski, MT Usuanlele, MJ Gilchrist, DJ Sharma, D Bulir, M Smieja, MG Surette, AG McArthur, D Mertz; Resources, J Gilchrist, DR Sharma, D Bulir, M Smieja, MG Surette, AG McArthur, D Mertz; Data Curation, SJC Baker, J Maciejewski, MT Usuanlele, J Gilchrist, D Bulir, M Smieja, AG McArthur, D Mertz; Writing – Original Draft, SJC Baker, M Smieja, AG McArthur, D Mertz; Writing – Review & Editing, J Maciejewski, MT Usuanlele, J Gilchrist, DR Sharma, D Bulir, M Smieja, MB Loeb, MG Surette, AG McArthur, D Mertz; Visualization, SJC Baker, D Mertz; Supervision, J Gilchrist, M Smieja, MB Loeb, MG Surette, AG McArthur, D Mertz; Project Administration, DR Sharma, D Bulir, M Smieja, AG McArthur, D Mertz; Funding Acquisition, M Smieja, MB Loeb, AG McArthur, D Mertz.

Ethics Approval:

Ethics approval for this research was obtained from the Hamilton Integrated Research Ethics Board (HiREB 4872-C).

Informed Consent:

N/A

Registry and the Registration No. of the Study/Trial:

N/A

Data Accessibility:

N/A

Funding:

This study was supported by the Hamilton Health Sciences Research Strategic Initiative Program (grant to DM). SJCB was funded by a MITACS Elevate Fellowship. AGM was supported by a David Braley Chair in Computational Biology.

Disclosures:

All authors report no conflicts of interest relevant to this article.

Peer Review:

This manuscript has been peer reviewed.

Animal Studies:

N/A