With the emergence of dd-cfDNA as a novel minimally invasive biomarker for allograft rejection supported by numerous clinical validation studies, a variety of analytical methods for dd-cfDNA quantification have been developed and utilized. However, besides the analytic validation, studies directly comparing different quantification approaches are lacking. Here we showed that ddPCR and HTS, the two most commonly used methods for dd-cfDNA quantification, are highly comparable and display only minimal bias for %dd-cfDNA. Moreover, this study represents to our knowledge the first development of an HTS method using UMI to enable absolute quantification of dd-cfDNA copies. Finally, urine was shown to contain substantial amounts of dd-cfDNA in stable kidney allograft recipients, albeit with a highly variable fractional abundance, supporting the requirement for absolute quantification to unlock the potential of urinary-derived dd-cfDNA as a fully non-invasive biomarker for renal graft transplant surveillance.
This HTS method incorporating UMIs for dd-cfDNA quantification without the need for donor-genotyping below 25% showed a high agreement of the %dd-cfDNA with ddPCR using assays targeting HLA-DRB1 alleles. The range of %dd-cfDNA measured in this study was between 0% and > 90% and included values greater than the range relevant in clinical diagnostics for plasma-based analyses, especially in hepatic and renal graft transplantation beyond the first month post-transplantation. However, as abnormal dd-cfDNA kinetics in the early post-transplantation phase with potentially high %dd-cfDNA of above 25% might be indicative of early adverse events in kidney transplant recipients such as urinary tract infections, pre-renal acute kidney injury or surgical complications among others , the accordance of both methods in a broad range is of great importance. Moreover, in the case of allogeneic hematopoietic cell transplantation, dd-cfDNA makes up the vast majority of cfDNA fragments  and therefore the relevant range is the opposite starting at 100%. Important to highlight is that these results show that even though both methods are based on completely different technologies and have different molecular targets, they result in a similar %dd-cfDNA. cfDNA release has been shown to be highly dependent on histone packaging and its susceptibility to DNase degradation, while also correlating with the transcription level of genes . As HTS methods interrogate multiple SNPs in the genome, any differences in the frequency of SNPs among the pool of (dd-)cfDNA fragments due to different cfDNA fragmentation would likely be canceled out. For ddPCR, on the other hand, with the assays used here targeting only a single locus in the HLA-DRB1 gene for the donor and recipient, such an effect is not expected. Moreover, even though the novel HTS method using UMI has not been as extensively validated as other HTS methods, we showed that a similar %dd-cfDNA could be measured when compared with AlloSeq® cfDNA, an analytically and clinically validated HTS assay. It is also worth mentioning that even though AlloSeq® cfDNA was validated for dd-cfDNA from plasma only, we also included urine samples in our analyses and still observed a high agreement between both methods. In contrast to the AlloSeq® cfDNA bioinformatic analysis, which corrected for close donor-recipient relatedness using constants, we implemented an EM algorithm that adjusts automatically for any skewing in the proportion of different donor SNP genotypes among recipient homozygous SNPs. The strong concordance of the %dd-cfDNA from both donor genotypes among informative SNPs corroborates the robustness of the algorithm. The algorithm can also be applied for a parent-child donor relationship with only heterozygous informative SNPs by defining the number of clusters as two instead of three.
To our knowledge, this is the first published incorporation of UMI to enable a more accurate absolute quantification of dd-cfDNA using HTS. The strong correlation with ddPCR suggests that HTS with UMI can be used to further evaluate the feasibility of HTS-based absolute dd-cfDNA quantification for minimally or non-invasive graft monitoring using dd-cfDNA. While the correlation with ddPCR was strong, a substantial proportional bias towards a lower number of dd-cfDNA copies detected by the HTS method was observed. This bias indicates that a lower proportion of input cfDNA molecules was sequenced with HTS compared to ddPCR based quantification, which could have different causes. First, there could have been a significant loss of fragments during ligation of adapters with the UMIs. Further, bead-based clean-up after adapter ligation before target enrichment might have substantially contributed to the loss of fragments. To utilize a method-independent diagnostic cutoff for absolute dd-cfDNA copy numbers, it is necessary to evaluate whether a sample-independent correction factor could be used to adjust absolute dd-cfDNA quantities from different methods. The use of such a correction factor could be feasible given that the ligation efficiency and fragment loss during clean-up are expected to be similar for all samples under the same conditions. Furthermore, the library preparation protocol has already been further improved by the manufacturer, replacing bead-based with enzymatic clean-up steps with the potential to minimize UMI loss after ligation and improving UMI ligation itself, which could increase the number of dd-cfDNA fragments in the library while potentially also minimizing variability.
We further applied ddPCR to determine and compare absolute and relative levels of dd-cfDNA in plasma and urine from stable liver and kidney transplant recipients. Plasma is the best-established source material for dd-cfDNA quantification so far and compared to urine, there is extensive literature on dd-cfDNA levels in plasma from allograft recipients. In 2017, the DART cohort-based study conducted by Bromberg et al.  investigated the median %dd-cfDNA and its variation in a reference kidney transplant population with the validated assay from CareDx. The median %dd-cfDNA of 0.19% (IQR: 0.01–0.43%) observed in this study is in accordance with the median of 0.21% (IQR: 0.12%-0.39%) in 380 blood samples reported by Bromberg et al. Similarly, Oellerich et al.  and Bloom et al.  reported median values of 0.29% (IQR: 0.17%-0.56%) and 0.30% (IQR: 0.14%-0.77%), respectively. For hepatic allograft recipients, Schütz et al.  conducted one of the most comprehensive studies in terms of the number of recipients included (n = 115). With a median %dd-cfDNA of 3.3% (95% CI 2.9%-3.7%) among stable liver transplant recipients, the level is similar even though a bit higher than the 2.2% (IQR: 0.72–4.1%) found in this study.
There is, however, limited data on the level of absolute copies in plasma from solid organ transplant recipients. The median number of dd-cfDNA copies/ml in kidney recipients with a negative biopsy in the cohort from Whitlam et al.  was 7 (IQR: 5–11), which was similar to the 5 copies/ml (IQR: 2–13) observed here. Oellerich and his colleagues  also investigated the absolute levels in stable kidney transplant recipients and found a median of 25 copies/mL (IQR: 11–60), which was five times higher than the number of dd-cfDNA copies detected here. One of the more prominent reasons for this discrepancy could be that both in our study as in that from Whitlam et al. no correction for the cfDNA extraction efficiency was used, while Oellerich et al. used an artificial spike-in to adjust for the extraction efficiency. Nevertheless, despite this difference in the mean dd-cfDNA copies observed, the inter-patient variation, expressed as the interquartile range relative to the median, was similar with 2.2 in our patient cohort and 2.0 in the study by Oellerich et al. Goh and colleagues  were to our knowledge the only ones who investigated absolute levels of dd-cfDNA in plasma from stable liver recipients and found a median of 66 copies/ml (IQR: 50–105) at day 42 post-transplantation in patients (n = 14) with an uneventful recovery. This is approximately half the dd-cfDNA copies/ml found here. However, it is noteworthy that the plasma samples from over half the stable liver recipients included here were collected one year post-transplantation and can therefore not be directly compared to the dd-cfDNA values at day 42. Further, Goh et al. may have underestimated the true number of copies as they did not adjust for non-amplifiable copies due to the amplicon length of the assays (50–130 bp) .
In contrast to plasma, urine would be a truly non-invasive source material with the additional benefit that it could be collected at home by the patients, and that it can easily be obtained in larger quantities compared to plasma. For this reason, we investigated the levels of dd-cfDNA in urine from solid organ recipients and found that the median %dd-cfDNA of 39.5% (IQR: 21.8–58.5%) in urine from stable kidney transplant recipients was slightly lower than the results revealed in other investigations such as from Lee et al.  with a mean of 53.3% (IQR: 21–92%) (n = 8) analyzed by sex-mismatch ddPCR and Burnham et al.  with a mean of 51.4% (n = 4) analyzed by HTS. With 30 samples from stable kidney recipients investigated here, the sample size was substantially larger in this study, which may explain the slight discrepancy to the two previous investigations based only on small patient numbers.
The only known study investigating urinary dd-cfDNA absolute copy numbers normalized by urinary creatinine levels so far was carried out by Sigdel et al. , reporting a mean number of 2.67 copies/µg creatinine. Based on a molar mass for creatinine of 113.12 gmol− 1, this corresponds to about 0.02 copies/µmol UCr, which is substantially lower than the 36.6 copies/µmol UCr measured in this study. A major reason for this discrepancy is that Sigdel et al.  log2 transformed the copies/ml before normalizing them by the urine creatinine levels. Applying the same transformation to the data in our study, the difference to the results from Sigdel et al. was strongly reduced with 5.2 copies/µmol UCr but remained substantial. A possible reason to explain the remaining difference might be that another method with a different extraction efficiency was used to extract the urinary cfDNA.
The results of the % dd-cfDNA in urine from stable kidney recipients revealed a large variability with values ranging from 0–80%, suggesting that urinary %dd-cfDNA is likely not a good biomarker candidate to identify or exclude kidney allograft rejections. This is supported by the observation that the extensive variation of the %ddcfDNA can predominantly be explained by the substantial fluctuations of recipient copies and less so by the donor copy variability. This large variation in recipient copies may potentially mask relevant changes in the amount of dd-cfDNA in urine if expressed relative to the total copies. Considering these results, the quantification of absolute dd-cfDNA levels will likely be needed to further evaluate the potential of urinary dd-cfDNA for allograft monitoring.
With the comparison of urine and plasma for the same stable allograft recipients, it was observed that for patients who received a kidney, both the %dd-cfDNA and absolute copies were higher in urine compared to plasma. One might thus hypothesize that more DNA is released from the kidney into the urine than into the blood circulation. A difference in cfDNA release dynamics is also supported by the lack of correlation of relative and absolute levels of dd-cfDNA between plasma and urine from the same patients. However, before any conclusions can be drawn, the comparison of the total amount of cfDNA released into the blood and urine would be needed to confirm this.
With our analysis of urine samples from stable liver transplant recipients, we aimed to gain further understanding about the phenomenon of transrenal dd-cfDNA, i.e. dd-cfDNA passing through the kidney into the urine. Even though a substantial number of dd-cfDNA copies could be detected in plasma in all liver recipients, interestingly, virtually no copies from the donor organ could be detected in urine from the same patients. These findings thus suggest very low levels of transrenal cfDNA in stable liver recipients. To our knowledge, there are no studies investigating relative nor absolute levels of transrenal dd-cfDNA originating from hepatic transplants. One study has investigated cfDNA in urine using methylation patterns specific for hepatocytes in healthy humans. Cheng et al.  detected virtually no transrenal cfDNA from hepatocytes, although no exact values were reported. Liver-derived cfDNA with just over 1%  made up the third-largest contributor to circulating cfDNA in plasma from healthy individuals besides blood- and vascular endothelium-derived cells. Based on this study and on our results, it is therefore unlikely that dd-cfDNA in urine could be used to monitor non-renal organ allografts with the methods currently available.
Of note, however, the extraction kit used for urinary-derived cfDNA had an isolation size range of 100 − 23’000 bp. Markus et al.  showed that the modal fragment size was 81 bp for all cfDNA fragments found in urine while the fragment size was 167 bp in plasma. It can therefore not be excluded that transrenal cfDNA predominantly consists of fragments significantly below 100 bp length. If transrenal cfDNA indeed has a shorter fragment length compared to plasma cfDNA, such fragments may have been lost during cfDNA isolation in our study. Biophysical studies on the length of transrenal dd-cfDNA in solid organ transplant recipients are lacking and results of studies on the length of transrenal cfDNA in other pathologies or pregnancy should not be directly translated, therefore the length of these fragments is still unknown. Alternative extraction kits or customized protocols should be investigated for the potential to extract short fragments so transrenal cfDNA might potentially be more efficiently isolated in solid organ transplant recipients. Comparison of kits with different degrees of short fragment extraction efficiencies would also reveal the extent of the effect that the extraction method has on the detection of transrenal DNA in urine.
Finally, this study has some limitations. First, a bit less than half of the 105 samples used for the ddPCR to HTS comparison were plasma samples. Therefore, body fluid-specific method agreement results are of limited statistical power. However, as both combined showed good agreement, big deviations are not expected if more samples were to be measured for each of the body fluids. A further limitation are the criteria, on which patients were deemed in a stable phase. For some kidney transplant recipients, a biopsy was not available and the decision was based on a low average serum creatinine, which has been shown to be an unsatisfactory biomarker to detect subclinical rejection. Similarly, the decision for the liver transplant patients was based on the levels of liver enzymes and total bilirubin, which show a low sensitivity and specificity in detecting subclinical acute cellular rejection. Since no liver biopsies were performed, no histological correlations could be shown. Therefore, it cannot be excluded that the data shown here also included patients with a subclinical rejection, which is associated with increased absolute and relative dd-cfDNA.