Use of next-generation sequencing to detect mutations associated with antiviral drug resistance in cytomegalovirus

ABSTRACT

Cytomegalovirus (CMV) is a significant cause of morbidity and mortality among immunocompromised hosts, including transplant recipients. Antiviral prophylaxis or treatment is used to reduce the incidence of CMV disease in this patient population; however, there is concern about increasing antiviral resistance. Detection of antiviral resistance in CMV was traditionally accomplished using Sanger sequencing of UL54 and UL97 genes, in which specific mutations may result in reduced antiviral activity. In this study, a novel next-generation sequencing (NGS) method was developed and validated to detect mutations in UL54/UL97 associated with antiviral resistance. Plasma samples (n = 27) submitted for antiviral resistance testing by Sanger sequencing were also analyzed using the NGS method. When compared to Sanger sequencing, the NGS assay demonstrated 100% (27/27) overall agreement for determining antiviral resistance/susceptibility and 88% (22/25) agreement at the level of resistance-associated mutations. The limit of detection of the NGS method was determined to be 500 IU/mL, and the lower threshold for detecting mutations associated with resistance was established at 15%. The NGS assay represents a novel laboratory tool that assists healthcare providers in treating patients who are infected with CMV harboring resistance-associated mutations and who may benefit from tailored antiviral therapy.

INTRODUCTION

Cytomegalovirus (CMV) infection is a common infection, with a seroprevalence of approximately 60% among adults in the United States (1). In immunocompetent individuals, CMV infection may be asymptomatic or result in mild, self-limited disease. Like the infection caused by other members of the Herpesviridae family, acute CMV infection is followed by lifelong latency (2). Periods of reactivation (i.e., lytic infection) may subsequently occur, which can go unrecognized in otherwise healthy individuals. However, in immunocompromised hosts (e.g., transplant recipients), primary or reactivated CMV can be associated with significant morbidity and mortality (3 – 5).

In patients with CMV disease, a number of antiviral medications are available to manage the infection, including ganciclovir, valganciclovir, foscarnet, cidofovir, and more recently, maribavir and letermovir (6). Among “high-risk” solid organ transplant (SOT) recipients [e.g., donor-seropositive (D+) and recipient-seronegative (R−) cases], valganciclovir prophylaxis has become a common strategy to prevent CMV disease. As more patients receive antiviral treatment to prevent or manage CMV disease, there has been an emergence of antiviral resistance, with studies describing CMV infections that are refractory to treatment (i.e., failure to demonstrate a reduction in CMV viral load of ≥1 log after 2 weeks of appropriate antiviral therapy) in up to 14% of high-risk SOT recipients (7). Resistance of CMV to antiviral medications is associated with specific mutations in several CMV genes, most notably UL97 and UL54, which encode a viral kinase and DNA polymerase, respectively (8). A number of mutations in UL97 confer resistance to ganciclovir, while specific mutations in UL54 may cause resistance to ganciclovir, cidofovir, and/or foscarnet (9).

Until recently, laboratory detection of CMV antiviral resistance has relied on Sanger sequencing of UL97 and UL54 (10 – 12). Although this has been a reliable approach, Sanger sequencing has several important limitations, including (i) poor sequence quality after ~700 base pairs, (ii) a threshold for variant detection of ≥20%, and (iii) a limited ability to detect and differentiate mixed sequence populations [e.g., majority wild-type (WT) sequence and minority mutant sequence in the same sample]. Due to these limitations, there is interest in the application of next-generation sequencing (NGS) to detect mutations associated with antiviral resistance in CMV. This technology may allow for earlier detection of resistance and the discovery of novel mutations associated with antiviral resistance.

In this study, we sought to develop a novel, NGS-based method to sequence the UL54 and UL97 genes of CMV and identify known mutations associated with antiviral resistance. The NGS assay was validated using the results of Sanger sequencing as the reference standard. We also describe the use of commercially available analysis software (DeepChek-CMV; ABL Diagnostics, SA; Woippy, France) that allows users to upload sequence files via a secured cloud-based portal and receive customized reports with amino acid mutations that are detected and predictions of antiviral susceptibility or resistance.

MATERIALS AND METHODS

Study design

Plasma samples (n = 27) submitted for either routine Sanger sequencing of UL54/UL97 genes at an outside reference laboratory or CMV viral load testing (cobas CMV; Roche Diagnostics, Indianapolis, IN) on the cobas 6,800 system (Roche Molecular Systems, Inc, Branchburg, NJ) were included in this study. The study protocol was reviewed and approved by the Institutional Review Board at Mayo Clinic.

Nucleic acid extraction and PCR amplification

Primers were designed to target the full-length UL54 and nearly full-length UL97 genes of CMV, yielding amplified products of 3,933 and 2,119 base pairs, respectively. The primer sequences were as follows: UL54 forward primer, 5′-AAG CTG TCA GCC TCT CAC GGT CC-3′, UL54 reverse primer, 5′-CGC GTC GCC GTT GCA CGT AG-3′; UL97 forward primer, 5′-TCC TCC GCA CTT CGG TCT CG-3′; UL97 reverse primer, 5′-TAC TCG GGG AAC AGT TGA CG-3′. For nucleic acid extraction, 1 mL of EDTA-plasma was pipetted into an easyMAG extraction cartridge (bioMérieux, Boston, MA) and placed on the NUCLISENS easyMAG (bioMérieux) for extraction according to the manufacturer’s protocol with an elution volume of 100 µL. Following extraction, 20 µL of the eluate was combined with 30 µL of LongAmp Hot Start Taq 2X master mix (New England BioLabs, Ipswich, MA) with a final primer concentration of 0.4 µM. Subsequently, a 96-well plate was placed on a Veriti instrument (Applied Biosystems, Foster City, CA) with thermocycling at the following parameters: 1 cycle at 94°C for 30 s; 45 cycles at 94°C for 30 s; 60°C for 30 s; 65°C for 4 min 15 s; 1 cycle at 65°C for 10 min; and 1 cycle at 4°C held indefinitely.

Following amplification, the 96-well plate was centrifuged for 10 sec on a PerfectSpin P plate centrifuge (PeqLab VWR, Radnor, PA). Subsequently, 5 µL of the amplified product was loaded into the well of a 1.2% FlashGel (Lonza Bioscience, Alpharetta, GA). Five microliters of the FlashGel molecular standard (100 bp to 4,000 bp) (Lonza) was added to a separate well, and the loaded gel was then placed in a FlashGel dock (Lonza) and run at 275 volts for ~5 min. The gel was then analyzed for PCR products of the appropriate size (UL97, ~2,000 bp; UL54, ~4,000 bp) using a FlashGel camera and gel capture software (Lonza). For each batch of clinical samples subjected to PCR, molecular-grade water (Roche, Basel, Switzerland) and WT CMV (Exact Diagnostics, Fort Worth, TX) without resistance-associated mutations in UL54/UL97 were included.

Next-generation sequencing

Following confirmation that PCR products of the appropriate size were present, 20 µL of the UL97 amplicon, 20 µL of the UL54 amplicon, and 20 µL of Tris-EDTA (TE) buffer (Thermo Fisher Scientific, Waltham, MA) were combined in a single well of a 96-well plate. Library preparation was performed using a TruSeq DNA Nano Library Prep Kit (Illumina, San Diego, CA) according to the manufacturer’s instructions. Briefly, mechanical DNA fragmentation was performed through ultrasonication (Covaris, Woburn, MA) to generate fragments of approximately 350 base pairs. DNA end repair and 3′ adenylation were performed, followed by ligation of index adaptors. DNA fragments were selectively enriched through PCR amplification using primers targeted to the index adaptor sequence prior to preparation for sequence analysis. Next-generation sequencing was performed using the MiSeq sequencer (Illumina) with 151-base pair paired-end reads or the HiSeq 2500 system (Illumina) with 101-base pair paired-end reads.

Sequence analysis and result interpretation

Following sequencing, the FASTQ file (i.e., a text-based format of nucleotide sequence and its corresponding quality metrics) was reviewed to ensure that the following quality control (QC) criteria were met: a total sample read count of ≥200,000; reads mapped to the reference sequence, ≥40,000 and ≥20% of total reads; mean position coverage, ≥40×; and mean variant coverage, ≥20×, which meets the New York State Department of Health guidelines (13). If the QC was acceptable, the FASTQ paired sequencing files were uploaded to the ABL website for analysis by DeepChek version 2.0 software (https://deepchek-mc.ablsa.com/), which compares the sequence to the reference database [human herpes 5 (Merlin) strain, complete genome NC_006273] and identifies UL54/UL97 mutations associated with resistance. DeepChek can also identify variants in other relevant genes, such as UL56, but this was not assessed during our study. Based on the verification experiments described later, the threshold for resistance reporting was set at 15%. Within 30 min of submission, a report providing predicted susceptibility or resistance to cidofovir, foscarnet, and ganciclovir, as well as specific mutations detected in UL54 and UL97, is provided (Fig. 1). The analysis software offers resistance interpretation based on peer-reviewed research literature (14), or from a CMV variant database that is updated as new data are available. Analysis for this study was performed using the CMV variant database updated in October 2018.

Fig 1.

ABL DeepChek-CMV drug resistance report summary. (A) Summary of CMV antiviral susceptibility (S) or resistance (R) prediction. Amino acid mutation analysis for (B) UL54 and (C) UL97. Coverage refers to the total number of reads (forward and reverse) at the position or the specific mutation. The ratio of reads for mutation and position coverage is used to determine prevalence. Mutations of interest (i.e., those previously shown to be associated with antiviral resistance) are shown in bold. Mutations with a prevalence of ≥15% are shown in red (i.e., these are reported clinically). Mutations with a prevalence of 5–15% are shown in blue (i.e., these are not reported).

Assay validation

In order to validate the performance of the CMV NGS assay, the following studies were completed: (i) precision (intra- and inter-assay), (ii) accuracy, (iii) reference range (i.e., normal value), (iv) analytical sensitivity [i.e., limit of detection (LoD)], and (v) analytical specificity.

Precision

Analyte-negative plasma samples were spiked with a plasmid containing the A594V mutation in the UL97 gene and a plasmid containing the N408K mutation in the UL54 gene (plasmids derived from the CMV laboratory strain AD169 provided by Dr. Sunwen Chou). Plasmids were quantified in copies/µL. To determine approximate concentrations in international units (IU)/mL, plasmids encoding UL54 were subsequently tested using the Roche cobas CMV assay. Spiking occurred at three different concentrations: 152,000 IU/mL, 18,300 IU/mL, or 1,990 IU/mL. Each of these samples was processed using the entire procedure described earlier, from nucleic acid extraction to sequence analysis. Samples were tested in triplicate on a single run (intra-assay) or in triplicate on three separate days (inter-assay).

Accuracy

A total of 43 plasma specimens were used to assess the accuracy of the assay. Analyte-negative plasma specimens (n = 16) were spiked with a plasmid containing a single resistance mutation in either UL97 (A594V, M460V, or C592G) or UL54 (D301N, N408K, A834P, or E756Q). Plasmid concentrations were tested on the NGS assay near the established limit of detection (500 IU/mL) and compared to expected results. Additionally, clinical plasma specimens (n = 27) submitted for CMV antiviral resistance testing by Sanger sequencing were compared to the results of the NGS assay.

Reference range

Plasma samples (n = 20) from normal donors underwent DNA extraction and PCR amplification using PCR primers for UL97 and UL54. None of the samples had detectable PCR products by agarose gel analysis, and therefore, they did not meet the criteria for subsequent testing by the NGS workflow.

Analytical sensitivity

To establish the limit of detection of the assay, three specimens were used: (i) an analyte-negative plasma spiked with wild-type CMV control (Exact Diagnostics), (ii) an analyte-negative plasma spiked with a plasmid harboring a known UL54 mutation (D301N), or (iii) a clinical specimen with a UL97 mutation (C603W) determined by Sanger sequencing. Specimens were diluted to achieve the following DNA concentrations: 2,000 IU/mL, 1,000 IU/mL, 500 IU/mL, 250 IU/mL, 50 IU/mL, 5 IU/mL, and 0.5 IU/mL. DNA extraction was performed on each dilution in triplicate, and the eluate from each extraction was tested in duplicate for a total of six replicates per concentration.

Analytical specificity

Plasma samples (n = 16) positive for HSV-1/2, VZV, EBV, HHV-6, BK virus, parvovirus, HBV, HCV, wild-type CMV, or human DNA underwent extraction and PCR amplification using UL97 and UL54 primers. The NGS assay was performed on any samples showing amplification product by agarose gel analysis. No CMV NGS result was generated for samples positive for viruses other than CMV. All CMV-positive plasma samples were reported as susceptible.

RESULTS

To assess the accuracy of the NGS assay, we compared the results of clinical samples previously tested by Sanger sequencing. Out of the 27 clinical samples that were tested, 59.3% (16/27) of the samples were reported as resistant by Sanger sequencing. The overall percentage of agreement between the assays was 100% (27/27) for resistance prediction (Table 1). Additionally, agreement was 88% (22/25) at the level of specific mutations reported. A comparison of the discrepant results is shown in Table 2. In specimens 1 and 2, additional UL97 variants were detected by NGS that were not detected by the Sanger sequencing method. The additional variants were detected at prevalences of 15.7% and 19.3%, respectively, which are below the 20% threshold required by the Sanger method.

TABLE 1.

Comparison of NGS and Sanger sequencing assays for the determination of antiviral drug resistance in clinical plasma samples

	Interpretation by Sanger sequencing
	Susceptible	Ganciclovir resistance	Cidofovir resistance	Foscarnet resistance
Interpretation by NGS
Susceptible	11	0	0	0
Ganciclovir resistance	0	16	0	0
Cidofovir resistance	0	0	1 a	0
Foscarnet resistance	0	0	0	2 b

^a Report for this sample also predicted ganciclovir resistance by both methods.

^b Report for these samples also predicted ganciclovir resistance by both methods.

TABLE 2.

Comparison of clinical specimens with discrepant results by the CMV NGS and Sanger sequencing assays ^c

Specimen	Sanger sequencing UL97	Sanger sequencing UL54	Sanger-predicted resistance	NGS UL97	NGS UL54	NGS-predicted resistance
1 a	H520Q	ND	G	H520Q C607Y	ND	G
2 a	H520Q	ND	G	H520Q M460V	ND	G
3 b	H520Q C607Y	ND	G	H520Q	ND	G

^a Multiple UL97 mutations (C607Y [15.7%], M460V [19.3%]) were detected by NGS analysis at a prevalence below the reporting threshold for Sanger sequencing (i.e., 20%).

^b UL97 mutation (C607Y) detected by Sanger sequencing was also detected by NGS, but at a prevalence of 9.02%. The predicted resistance of these specimens was the same by both assays.

^c Mutations present at a prevalence of ≥15% were considered significant and indicated as present. TNP, test not performed; ND, not detected; G, ganciclovir; F, foscarnet.

To further study the accuracy of the NGS method, plasmids harboring known mutations in UL54 or UL97 were spiked into analyte-negative plasma specimens and tested by the NGS assay. The assay correctly identified the mutation and associated resistance pattern in 14 (87.5%) of 16 spiked plasma samples (Table 3). Of note, an additional UL54 mutation (A834P) was detected by NGS on initial testing for two samples (Table 3, specimens 15 and 16) at a prevalence of 17.2% and 15.9%, respectively; however, this mutation was not detected by NGS upon repeat testing, yielding an overall agreement of 100% (16/16).

TABLE 3.

Summary of spiking studies with plasmids containing mutation(s) known to confer resistance ^{a ,b,c}

Specimen	Spiked mutation	UL54 mutation detected	UL97 mutation detected	Predicted resistance
1	A594V	ND	A594V	G
2	C592G	ND	C592G	G
3	M460V	ND	M460V	G
4	A594V	ND	A594V	G
5	C592G	ND	C592G	G
6	M460V	ND	M460V	G
7	A834P	A834P	ND	G/C/F
8	A834P	A834P	ND	G/C/F
9	D301N	D301N	ND	G/C
10	D301N	D301N	ND	G/C
11	N408K	N408K	ND	G/C
12	N408K	N408K	ND	G/C
13	E756Q	E756Q	ND	F
14	E756Q	E756Q	ND	F
15	E756Q	E756Q/A834P	ND	G/C/F
15 RPT	E756Q	E756Q	ND	F
16	E756Q	E756Q/A834P	ND	F
16 RPT	E756Q	E756Q	ND	F

^a Samples were spiked near the limit of detection.

^b Plasmids were derived from the CMV laboratory strain AD169.

^c ND, not detected; RPT, repeat; G, ganciclovir; C, cidofovir; F, foscarnet.

Precision testing was performed to evaluate the intra- and inter-assay reproducibilities of the NGS assay. Plasmids containing a mutation in UL97 (A594V) or UL54 (N408K) were used to spike analyte-negative plasma at three different concentrations and tested in triplicate in the same run (i.e., intra-assay precision). Testing was repeated for a total of three days (i.e., inter-assay precision), and all reported results demonstrated 100% agreement with the expected result. Additionally, the percentage of prevalence reported remained consistent across testing days. The range of values for mutation prevalence was within 2% for all concentrations and mutations tested.

The limit of detection (LoD) of the assay was assessed by testing a dilution panel of plasma samples known to be positive for CMV harboring either a UL97 or a UL54 resistance-associated mutation. Each member of the dilution panel was tested in six replicates, and the highest dilution (i.e., lowest concentration) determined to be positive for the expected mutation and overall resistance prediction was categorized as the LoD. For UL97, all quality metrics passed, and the expected mutation (C603W) and resistance prediction were accurate in all six replicates at a concentration of 500 IU/mL. For UL54, all quality metrics passed, and the expected mutation (D301N) and resistance prediction were accurate in all six replicates at a concentration of 250 IU/mL.

To determine whether the NGS assay could detect multiple variant populations in the same sample, plasmids harboring mutations in UL54 or UL97 were used to spike at various concentrations into analyte-negative plasma. Three combinations were tested: (i) a sample spiked with three unique UL97 mutations, (ii) a sample spiked with three unique UL54 mutations, and (iii) a sample spiked with 3 UL97 and 3 UL54 mutations (Table 4). All expected mutations were detected in samples spiked with multiple mutations. To determine whether the NGS assay can improve the ability to detect low-level variants among mixed sequence populations, plasmids containing wild-type UL54 or UL54 with the D301N variant were spiked into analyte-negative plasma at varying ratios (Table 5). The D301N variant was detected at the lowest ratio tested (90% WT:10% variant); however, an unexpected resistance-associated mutation (A594T) was detected at a threshold of 14.2% in one replicate of the WT control, and therefore, the threshold for variant prevalence of the assay was established at 15%.

TABLE 4.

Detection and differentiation of combinations of UL97 and UL54 mutations by the CMV NGS assay ^a

Mutation combinations in spiked samples		Predicted resistance (Antiviral[s])	UL97 mutations detected (% prevalence)	UL54 mutations detected (% prevalence)
UL97	M460V	Resistant (G)	M460V (34.4%)	None detected
	C592G		C592G (28.5%)
	A594V		A594V (36.5%)
UL54	D301N	Resistant (G, C, F)	None detected	D301N (23.5%)
	N408K			N408K (49.5%)
	A834P			A834P (25.0%)
UL97	M460V	Resistant (G, C, F)	M460V (34.9%) C592G (27.3%) A594V (35.6%)	D301N (27.0%) N408K (49.8% A834P (23.1%)
	C592G
	A594V
UL54	D301N
	N408K
	A834P

^a Mutations present at a prevalence of ≥15% were considered significant and indicated as present. G, ganciclovir; C, cidofovir; F, foscarnet.

TABLE 5.

Summary of mutation detection in a mixed population of a UL54 mutant and wild-type CMV ^a

Ratio (%) of wild-type:UL54 mutant	UL54 mutation	Predicted resistance
50:50	D301N	G/C
75:25	D301N	G/C
80:20	D301N	G/C
85:15	D301N	G/C
90:10	D301N	G/C

^a Mutations present at a prevalence of ≥15% were considered significant and indicated as present. G, ganciclovir; C, cidofovir.

DISCUSSION

There are a growing number of antiviral agents used for prophylaxis and treatment of CMV infection. However, there is a trend of increasing CMV resistance to one or multiple available antivirals. Sanger sequencing has been the traditional method used to detect antiviral resistance, but this method has a limited ability to detect and differentiate mixed populations and generally requires a prevalence of ≥20% for variant detection. The use of NGS technology has been shown to improve the detection of CMV resistance and may identify emerging resistance earlier in the disease course (11).

In our study, the NGS assay and Sanger sequencing were found to predict identical resistance for 100% (27/27) of clinical samples tested. This demonstrates that the NGS method represents a reliable alternative for CMV antiviral resistance determination. However, we observed 88% (22/25) agreement between Sanger and NGS assays at the level of detecting specific resistance-associated mutations. The three discrepancies were limited to three patient samples. The three samples showed an H520Q mutation in UL97 (predicting ganciclovir resistance) by both methods, but a unique mutation(s) was(were) also identified by either NGS or Sanger assay in each of these specimens (Table 2). In specimens 1 and 2, NGS also identified a C607Y and an M460V mutation at a prevalence of 15.7% and 19.3%, respectively, which were not detected by the Sanger assay. The presence of these additional mutations did not alter the overall resistance prediction (i.e., ganciclovir resistance). In specimen 3, the Sanger assay identified the presence of the C607Y UL97 mutation, which was not reported by the NGS assay; however, an analysis of the NGS data showed the C607Y mutation to be present at a prevalence of 9.02% (Table 2).

Additionally, our work and others have found that NGS-based methods may lower the threshold for detecting resistance-associated mutations compared to Sanger sequencing (15, 16). The results from our mixed population studies with WT and resistance-associated mutations suggested that mutations could be detected at a variant concentration of 10% (Table 5). However, during the assay validation studies, there was an isolated occurrence of a resistance-associated mutation being detected in the WT CMV control at a prevalence of 14.2%. For this reason, the reporting threshold was established at 15% when the assay was implemented in the clinical laboratory. It is likely that some of the discrepancies between the two methods are due to the improved detection of low-prevalence mutations by the NGS assay. While clinical correlation of low-prevalence mutations will be essential, the lower detection threshold of NGS may allow for earlier identification of resistance or emerging resistance in some patients. Additionally, detection of low-prevalence mutations (i.e., 15–20%) by the NGS method may warrant additional testing and/or close monitoring of the patient’s response to antiviral therapy.

NGS-based assays are associated with higher costs than Sanger sequencing. However, it is expected that the cost will decrease with further development of the technology and more widespread use in clinical laboratories. The increased cost of NGS may be justified by our data and others, which have demonstrated that NGS lowers the prevalence threshold required for resistance detection and has better performance with mixed genetic populations. The use of NGS can potentially provide information about resistance-associated mutations, which can ultimately affect treatment decisions sooner than Sanger sequencing.

This study has several limitations. First, we were able to compare the results of NGS and Sanger assays in only 27 patients. However, we supplemented the assessment of accuracy by performing spiking studies on an additional 16 plasma specimens, which demonstrated 100% agreement between the methods. Second, discrepancies between NGS and Sanger assays could not be further characterized due to insufficient sample volume and the lack of correlation to clinical outcomes due to specimens being submitted through our reference laboratory.

This report summarizes the development and performance verification of a novel NGS-based method for the detection of antiviral resistance in patients with CMV infection. Further work is needed to continue to understand the utility of this assay, assess its ability to identify novel resistance-associated mutations, and study the kinetics of emerging resistance. Additionally, future studies should focus on the clinical significance of low-prevalence mutations (i.e., those detected by NGS at a prevalence between 5 and 15%). Investigating patients with low-prevalence mutations and how they respond to treatment over time will be important to define the potential significance of minor variant populations. In summary, we have developed a novel NGS method that can reliably and accurately predict resistance to common antivirals, including ganciclovir, cidofovir, and foscarnet. This method will assist in the timely management of patients who are treated for CMV infection with clinical and/or laboratory evidence of antiviral resistance.