2.1 Clinical specimens and patients

In this retrospective observational study, a total of 52 plasma specimens with detectable CMV DNAemia from 14 nonconsecutive randomly selected allo-SCT recipients, collected between May 2020 and May 2022, were included (Table S1). We collected 16 plasma samples (median, four specimens/patient; range, 1−7) from patients on LMV that developed breakthrough CMV DNAemia (median time after LMV inception, 23 days; IQR, 18−29) (LMV group). All episodes of LMV-breakthrough CMV DNAemia cleared spontaneously. A total of 23 plasma specimens were available from 10 LMV-untreated patients with CMV DNAemia who were not receiving PET at the time of sampling (non-PET group). Finally, a total of 13 specimens from five LMV-untreated patients who developed clinically significant CMV infection (CsCMVi), that is, requiring PET as per local guidelines, were also available (median three specimens/patient; range 3−4). These 13 specimens were collected while patients were under (val)ganciclovir treatment (Under-PET group). The current study was approved by the institutional ethical review board of the INCLIVA Biomedical Research Institute (January 2020). The request for informed consent was waived by the institution due to its retrospective nature.

2.2 Monitoring and management of CMV infection

The RealTime CMV PCR (Abbott Molecular), a dual target (UL34 and UL80.5) PCR, with a limit of detection and quantification of 31.4 IU/mL (95% confidence interval) was employed to monitor CMV DNA load in plasma. Amplification reactions were run on the m2000rt platform (Abbott). Viral DNA extraction was performed using the Abbott m2000sp platform (Abbott). LMV prophylaxis was administered to high-risk CMV seropositive patients at conventional doses within the week after cell infusion until day +100, following local guidelines. PET with valganciclovir (900 mg/12 h) or intravenous ganciclovir (5 mg/kg/12 h) was initiated, regardless of whether the patients were under LMV prophylaxis, when the plasma CMV DNA load reached ≥ 1500 IU/mL, or when the CMV DNA doubling time was ≤2.0 days, whichever occurred first.^{21, 22}

2.3 Assessment of CMV DNA fragmentation

Experiments assessing CMV DNA fragmentation in plasma were carried out in two phases. In the first phase, which took place before the implementation of LMV at our center, we investigated whether the CMV DNA fragmentation profile may differ across or within five randomly selected allo-SCT recipients. A total of 45 end-point PCR assays targeting the CMV UL34 (n = 20), UL80.5 (n = 19), and UL54 (n = 6) genes were used in preliminary experiments, as described below. The primer pairs (Table S2) were designed to overlap with the neighboring ones and to amplify fragments of 90−110 bp to cover the corresponding full-length gene sequences of UL34 and UL80.5 (1224 and 1122 bp, respectively), and a selected region of 368 bp of UL54, UL34, UL80.5, and UL54 sequences were cloned into pGEMTeasy vectors to generate plasmids incorporating the corresponding sequence. These plasmids were used as templates for the amplification reactions. All PCR reactions generated amplicons with the expected size (data not shown). In the second phase of the study, after the implementation of LMV at our center, quantitative real-time PCR assays generating amplicons that were differentially detected in study phase one (primer pairs; UL34_4, UL34_12, UL34_17, UL34_18, UL54_3, UL54_5, UL80.5_2, UL80.5_5, UL80.5_6, UL80.5_12, and UL80.5_19), were developed using an Exicycler RT-PCR thermal cycler (Bioneer), RT2 SYBR green Rox qPCR Mastermix (Qiagen), following the manufacturer's instructions. Each 10 μL reaction contained 5 μL of x2 Master Mix (Qiagen), 1.5 μL of each primer (0.25 μM final concentration each), and 2 μL of eluate. A positive CMV control and a negative control (nuclease-free water) were included in every PCR run. Serial dilutions of the plasmid containing the respective CMV gene (from 10² to 10⁸ genome equivalent/mL) were used to generate a calibration curve (genome copies number/μL) and to estimate primer pair efficiencies by using the calibration dilution curve and slope calculation method (Efficiency (%) = (10^{(−1/slope)-}1) x 100).²³ The cycling conditions were the following: initial denaturation at 95°C for 5 min, 45 cycles of 95°C for 15 s, and 60°C for 1 min, followed by a melting curve ranging from 60°C to 95°C (acquiring fluorescence data every 0.3°C).

2.4 Statistical analysis

A comparison of categorical variables was carried out with the Chi-Square test or Fisher's test, as appropriate. Comparison of continuous variables (medians and IQRs) between two or more groups was carried out with the non-parametric Mann−Whitney U test or the Kruskal−Wallis test, respectively. Correlation between continuous variables was assessed with the Spearman Rank test. Based upon the quantification of specific PCR amplicons, a logistic regression analysis was performed to generate a model that would categorize plasma specimens as belonging to LMV or non-LMV-treated patients. All amplicons (in genomes/µL) were introduced as quantitative variables into the model and discarded one by one by the backward-elimination rule until the remaining variables reached statistical significance. Performance metrics included accuracy, area under a curve (AUC), sensitivity, specificity, precision, and F-measure. The diagnostic performance was assessed by a confusion matrix. p ≤ 0.05 (two-tailed) were considered statistically significant. Graphs and statistical analyses were performed using GraphPad Prism software version 9.0.

3.1 Intra- and inter-patient qualitative differences in the CMV DNA fragmentation pattern in plasma

In the second phase of the study, we optimized quantitative real-time PCR reactions using the aforementioned selected primer pairs for further experiments. We investigated whether the CMV DNA fragmentation profile as inferred from the number of detected amplicons differed within a given patient over time, between patients belonging to the same study group, or across patient groups (Table 1). Overall, one or more PCR amplicons were not detected in 29/52 samples (56%). Specimens with missing PCR amplicons were unevenly distributed across study groups: 14/16 (88%) in samples from LMV patients, 11/16 (69%) in samples from non-PET patients, and 4/20 (20%) in samples from under-PET patients (p < 0.01). Nevertheless, the mean number of missing PCR amplicons was not significantly different across groups (2, 1, and 1 in LMV, non-PET, and under-PET patients, respectively; p > 0.05). PCR amplicons that were differentially detected included those generated by primer pairs UL34_17, UL54_3, UL54_5, UL80.5_2, UL80.5_5, UL80.5_6, and most notably UL80.5_19. Moreover, as shown in Figure 2, intra-patient variations in the number of detected PCR amplicons over time were observed (in 3/3 LMV patients sampled more than once, 5/5 non-PET patients, and 3/5 Under-PET patients).

As stated above, specimens with missing PCR amplicons were substantially less represented among under-PET plasma specimens compared with non-PET and LMV samples. Since CMV DNA loads measured by the Abbott assay were overall higher in under-PET specimens, we next investigated whether the number of detected amplicons was directly related to the magnitude of the CMV DNA load. No such association was evident (overall data in Table 1 and Figure 3). We conducted a series of subanalyses to further assess this issue. First, we compared the number and profile of PCR amplicons generated from specimens from LMV and non-PET patients displaying CMV DNA loads < 500 IU/mL (median, 127 IU/mL [IQR, 78−193] and 431 IU/mL [IQR, 372−463], in LMV-patients and non-PET patients, respectively). As shown in Figure 4A, the frequency of specimens with 10 or 11 PCR amplicons detected was higher (p = 0.04) among LMV patients (57%) compared with that in non-PET patients (38%). Moreover, the PCR amplicon profile was substantially different between LMV samples and non-PET specimens (Figure 4B); notably, UL54 PCR amplicons were more frequently missing in specimens from LMV patients than in those from under-PET patients, whereas the opposite held true for UL80.5 PCR amplicons, especially UL80.5_19 (Figure 4B). Likewise, evident differences in both the number and pattern of PCR amplicons detected were noticed across non-PET and under-PET specimens, irrespective of the CMV DNA load, graded as between 500 and 5000 IU/mL (median 984 IU/mL [IQR 541−1829] in non-PET specimens and median, 2,345 IU/mL [IQR 1396−3805], in Under-PET specimens), or >5000 IU/mL (median, 7392 IU/mL [IQR 6943−13 975] in non-PET samples and median, 10 962 IU/mL [IQR 8094−19 602], in Under-PET specimens) (Figures 4C,D). In this sense, strikingly, among specimens with high viral loads (>5000 IU/mL), those from under-PET patients had fewer detected PCR amplicons (40% of specimens with 11 detectable amplicons) than those from non-PET patients (100% of specimens with 11 amplicons), despite higher CMV DNA loads. Importantly, no PCR amplicon profile could consistently discriminate between patient groups.

We also assessed whether the efficiency of the different real-time PCR reactions had an impact on the pattern of PCR amplicons observed across specimens. The efficiency of real-time PCR assays generating amplicons differentially detected across specimens varied between 61% (UL80.5.6) and 109% (UL80.5.2) (data shown in Table 1). We observed no obvious association between the above two parameters; for example, in specimens from LMV and non-PET (CMV DNA load, <500 IU/mL), the UL80.5.19 amplicon was substantially more frequently undetected than UL80.5.5, despite PCR reactions displaying a comparable efficiency (79% vs. 72%) (see Figure 4B).

3.2 Differences in CMV DNA loads quantified by amplicon-specific PCR assays across patients

We next aimed to determine whether there were differences in CMV DNA loads measured by UL34, UL80.5, and UL54 real-time PCR assays (n = 11) across specimens from patients in the different study groups. We first noticed that the correlation between CMV DNA loads quantified by the Abbott RealTime CMV PCR assay (in IU/mL) and those measured by the different “in-house” real-time PCR assays (in genome copies/µL) varied widely: the correlation was strong (rho, 0.61−0.80; p < 0.001) for UL54_3, UL34_18, and UL80.5_12 real-time PCR assays, moderate (rho, 0.41−0.60; p < 0.001) for UL34_12, UL34_17, UL80.5_5, UL80.5_6, and UL80.5_19, and absent for UL34_4, UL54_5, and UL80.5_2. The level of correlation was seemingly unrelated to the respective efficiency of the real-time PCR assays (not shown).

Relevant differences in CMV DNA loads quantified by laboratory-developed real-time PCR assays were as follows (Table 2): (i) CMV DNA loads measured by the UL34_4 and UL80.5_2 real-time PCRs were significantly higher in LMV specimens than in non-PET specimens displaying CMV DNA loads <500 IU/mL in the Abbott assay (p < 0.001 and p = 0.007, respectively). In contrast, CMV DNA load quantified by the UL34_18 real-time PCR assay was significantly lower (p = 0.04) in specimens from the LMV patients as compared with those from non-PET specimens; (ii) Only CMV DNA load as measured by the UL54_3 real-time PCR assay differed significantly between non-PET and Under-PET specimens (p = 0.04 for CMV DNA loads between 500 and 5000 IU/mL, as measured by the Abbott assay) (Table 2).

3.3 CMV DNA loads quantified by UL34_4, UL54.5, and UL80.5_1 PCRs discriminate between plasma from LMV and non-LMV patients

A stepwise logistic regression was performed using quantitative data from all 11 UL34, UL80.5, and UL54 real-time PCR assays to build a model that was tested for its ability to discriminate between specimens from patients undergoing LMV prophylaxis and those from LMV-untreated patients (including non-PET and Under-PET). The data are shown in Figure S1. While quantitative data from any individual real-time PCR assay did not allow discrimination, the combination of specific cut-offs selected by the model (>20 genome copies/µL of UL34_4, >500 genome copies/µL of UL54.5, and <5 genome copies/µL of UL80.5_12) afforded perfect discrimination (accuracy, AUC, sensitivity, specificity, and precision of 100%).