Human cytomegalovirus (HCMV) is a ubiquitous betaherpesvirus with significant disease-causing potential in immunocompromised patients, including children with congenital immune deficiencies or immune suppression following solid organ or bone marrow transplantation. As well as causing pneumonitis, colitis, retinitis, and uveitis (Rafailidis et al., 2008) all of which contribute to HCMV-related mortality, HCMV disease also increases the risk of allograft vasculopathy and graft rejection, and significantly increases treatment costs (Hiwarkar et al., 2013). Children are at particular risk from HCMV, with over 25% of primary HCMV infections in the UK occurring in childhood (Patrick et al., 2014). Up to 16% of patients on prolonged anti-HCMV therapy develop drug resistance (Couzi et al., 2012; Shmueli et al., 2014), many of them with mutations which cause multi-drug resistance (Hantz et al., 2010). However, it may be that not all the mutations that cause resistance are known and this may lead to underestimation of drug resistance in patients failing therapy.

Three drugs are currently licensed for HCMV prophylaxis and treatment, including ganciclovir (GCV), foscarnet (FOS), cidofovir (CDV); brincidofovir (the oral derivative of cidofovir), and letermovir are in phase III clinical trials; maribavir (MBV) is available on a compassionate use basis. Treatment failure occurs in between 20 (Asberg et al., 2007) and 50% (van der Beek et al., 2012) of HCMV cases, necessitating drug changes and in some cases the use of adoptive immunotherapy. Genetic evidence of drug resistance can guide clinical decision making (Houldcroft, 2015) but current methods have technical limitations. Sanger sequencing of PCR amplicons only reliably detects drug resistance mutations that are present at frequencies of 20% or more (Sahoo et al., 2013). Deep sequencing of PCR amplicons has enabled detection of minority resistance variants at frequencies as low as 1% (Görzer et al., 2010) which could lead to earlier detection of HCMV resistance and better treatment. However, PCR and nested PCR are known to generate mutations which could make the identification of low level resistance mutations more difficult (Depledge et al., 2011). To minimize this problem, and to capture the genes currently implicated in antiviral resistance simultaneously, we made use of novel target enrichment (Depledge et al., 2011) and deep sequencing to analyse the UL27, 54, and 97 genes in serial samples from patients with prolonged HCMV viraemia despite anti-HCMV therapy. In this study, we include 11 retrospectively identified patients from Great Ormond Street Hospital for Children who had high HCMV loads for 2 weeks or longer, with clinician suspicion of anti-viral drug resistance.

The duration of HCMV positivity and treatment for each of the 11 patients is shown in Table 1. Using SureSelect target enrichment we recovered sequence mapping to the UL27, 54, and 97 genes directly from all the clinical diagnostic samples in a single reaction without the need for virus isolation or PCR of overlapping genome fragments. A sample read mapping plot for each ORF is shown in Supplementary Figure 1. Details of mapping and coverage relative to virus genome copies/ml blood are shown in Supplementary Table 1 and Supplementary Figure 2.

From deep sequencing results we were able to stratify patients into two groups: those with no evidence of developing resistance mutations despite receiving long term antiviral treatment (A, C, D, G, J, K, L); and those patients who developed known HCMV resistance mutations: either fixed (H and M) or at low level (B and I). We plotted viral load, drugs received and mutations over time for each of these patients (Figure 1; Supplementary Figure 3).

Patients I and L (despite never having received MBV) showed evidence of fixed truncating mutations in UL27 (Supplementary Figure 4A) both of which would be predicted to confer resistance and/or growth attenuation (Chou, 2009; Hakki et al., 2011). In patient G, a minority stop codon (~10%) was detected at amino acid position 512 in UL54 day 63 post-admission, but was not detected in subsequent samples from this patient (Supplementary Figure 4B).

In samples from a number of patients, we detected low-frequency frame shift mutations in ORF UL54, at frequencies of between 2 and 13%: A (< 5%); B (< 10%); C (< 6%); D (< 12%); H (10%); and K (13%) (Supplementary Figure 5). Many of these mutations were lost over time, or replaced by different frame shifts, suggesting they are unfit.

Persistent HCMV viraemia is associated with poor outcomes in immunosuppressed patients, including those undergoing bone marrow (Hiwarkar et al., 2013) and solid organ transplantation, and treatment with anti-HCMV drugs is indicated. HCMV viraemia carries significant economic costs, estimated at £22,500 ($32,000) per pediatric bone marrow transplant patient (Hiwarkar et al., 2013). To explore treatment failure, testing for resistance mutations and if necessary a change in therapy is recommended if the viral load remains the same or rises after 2 (Boeckh and Ljungman, 2009) or 3 (van der Beek et al., 2012) weeks of treatment. Changes in treatment may also be prompted by side effects, and bone marrow function particularly in hematological stem cell transplant recipients. In this study we used deep sequencing to investigate drug resistance patterns in persistently viremic patients requiring prolonged treatment. Notwithstanding persistent viraemia, seven patients showed no sign of drug resistance and six of them were able to control their viremia to below 10³ gc/ml while on treatment. Patients who developed resistance had higher viremia, lower lymphocyte counts, more drugs and longer duration of antiviral treatment, although numbers were too small for these differences to be significant. Overall, these data support the findings of others, that development of drug resistance mutations are associated with poor control of viremia and represent a poor prognostic indicator in immunosuppressed patients receiving treatment for HCMV; two of four patients developing resistance mutations died as compared with one of seven who remained resistance-free. Notwithstanding these findings, the two patients H and M, in whom resistance mutations rose rapidly to fixation, responded to a change in treatment and controlled their viremia (two qPCR results < 10³ gc/ml) within a mean of 17 weeks. In patient M the H520Q resistance mutation to GCV in ORF UL97 persisted despite withdrawal of the drug, suggesting that this variant remained fit despite the H520Q mutation.

By contrast, where we identified multiple mutations occurring simultaneously, in patients B and I, this was associated with profound treatment failure and death from HCMV-related disease. Observations from deep-sequencing of PCR amplicons suggest that multiple resistance mutations occurring at sub-fixation levels can contribute to a drug-resistant phenotype (Chou et al., 2014) and this is consistent with the evidence, particularly in patient B in whom high HCMV viral loads persisted in the presence of multiple often low frequency mutations (Figure 1). One explanation is that low frequency drug resistance mutations are distributed throughout the viral population resulting in many relatively unfit resistant viruses, none of which can outcompete the others (Chou et al., 2014). A similar pattern in seen in patient I, in whom a change from GCV to CDV appears to have selected for one set of resistance mutations in favor of another, perhaps because these mutations arose on different populations of the virus within this patient. Further evidence for this comes from mouse studies making use of cells infected with multiple murine CMV strains. These strains trans-complement one another, increasing overall viral fitness (Cicin-Sain et al., 2005). Both patients I and B showed a rapid rise in resistance mutations in response to treatment changes, with concomitant loss of others. This pattern, particularly in patient B for whom more samples were available, is consistent with low level persistence of multiply resistant viruses which rapidly replicate under the selective pressure of a new drug. Conventional PCR and Sanger sequencing failed on at least two occasions to detect any of these mutations, apart from the D588N which was present at a frequency of 9%. Deep sequencing is therefore able to detect potentially important drug resistance that is missed by conventional methods. For example the P552Q mutation was detected at frequencies of 1.67% (day 119), 4% (day 126), and 10.64% (day 133), prior to the start of MBV on day 133. Similarly, PCR and Sanger sequencing of samples from patient I missed multiple drug resistance mutations at frequencies of 2–41%, 54 days after these mutations became detectable by target-enriched sequencing.

The speed with which the virus became resistant in patient B and the rapid loss of four drug resistance mutations in UL97 and UL54, suggested strain replacement rather than de novo mutation and prompted us to examine the possibility of mixed infection. The change in phylogenetic clustering for UL27 sequences following the introduction of MBV confirmed this suspicion. HCMV is known to be highly recombigenic (Sijmons et al., 2015), and in this case, without whole genome analysis, we are unable to distinguish the possibility of recombination, re-infection, or reactivation of a pre-existing secondary strain.

In summary we have demonstrated that deep sequencing of HCMV ORFs UL27, 54, and 97 could be achieved directly from whole blood with virus loads in the range 80,000–65,000,000 copies/ml without prior culture or PCR. We were able to detect resistance mutations occurring at 2% or more in patients with viraemia persisting at levels of ≥10⁴ gc/ml for 2 weeks or more, comparable to other deep sequencing-based approaches (e.g., Görzer et al., 2010; Garrigue et al., 2016). Our data suggest that in contrast to amplicon and Sanger sequencing, our deep sequencing can exclude resistance in patients with persistently high levels of viraemia, thereby providing a measure of support prolonging current antiviral treatment or returning to them at a later date if further treatment is needed. Where resistance mutations are detected, we observed two patterns, rapid development of fixed resistance with clearance of virus following a change in treatment, and development of multiple sub-fixation resistance mutations, with potentially poorer outcome. Further investigation is needed to determine whether these patterns are indeed predictive of outcome. We do not yet understand why multiple minority drug resistance mutations arise in some patients. Multiple minority variants, which are likely to be better detected using deep sequencing methods, appeared to complicate treatment to a greater extent than single fixed resistance mutations. In our patients multiple low level drug mutations was associated with poor prognosis probably because they increased the risk that a change in drug would select for a pre-existing mutation. Deep-sequencing of HCMV allows us to characterize these mutations and could be used to inform which drugs are given earlier in treatment, or to highlight those patients for whom additional non-pharmacological interventions such as withdrawal of immunosuppression, or the use of virus-specific cytotoxic T lymphocytes are most appropriate.

Raw sequencing data has been deposited in the European Nucleotide Archive under project accession PRJEB12814. Bait sequences are available by request from the authors.

JB, CH, and DD conceived the study design. CH, EW, JS, AW, SM, and PV supplied patient clinical data and interpretation. CH, DD, and SN performed the DNA extractions. DD, HT, CH, and SN sequenced the samples. RW administered the study. CH, JMB, and DD analyzed the data. CH, JMB, and JB wrote the paper. All authors read and approved the manuscript.

This work was supported by funding from the European Union's Seventh Programme for research, technological development and demonstration under grant agreement no. 304875. JB is supported by the UCL/UCLH and CH by the UCL/GOSH Biomedical Resource centers. DD is supported by an MRF New Investigator Award. SN was funded by a Microbiology Society Harry Smith vacation scholarship. JB receives funding from the UCL/UCLH NIHR biomedical research center.