Deep sequencing of DNA from urine of kidney allograft recipients to estimate donor/recipient-specific DNA fractions

Abstract

Kidney transplantation is the treatment of choice for patients with end-stage kidney failure, but transplanted allograft could be affected by viral and bacterial infections and by immune rejection. The standard test for the diagnosis of acute pathologies in kidney transplants is kidney biopsy. However, noninvasive tests would be desirable. Various methods using different techniques have been developed by the transplantation community. But these methods require improvements. We present here a cost-effective method for kidney rejection diagnosis that estimates donor/recipient-specific DNA fraction in recipient urine by sequencing urinary cell DNA. We hypothesized that in the no-pathology stage, the largest tissue types present in recipient urine are donor kidney cells, and in case of rejection, a larger number of recipient immune cells would be observed. Extensive in-silico simulation was used to tune the sequencing parameters: number of variants and depth of coverage. Sequencing of DNA mixture from 2 healthy individuals showed the method is highly predictive (maximum error < 0.04). We then demonstrated the insignificant impact of familial relationship and ethnicity using an in-house and public database. Lastly, we performed deep DNA sequencing of urinary cell pellets from 32 biopsy-matched samples representing two pathology groups: acute rejection (AR, 11 samples) and acute tubular injury (ATI, 12 samples) and 9 samples with no pathology. We found a significant association between the donor/recipient-specific DNA fraction in the two pathology groups compared to no pathology (P = 0.0064 for AR and P = 0.026 for ATI). We conclude that deep DNA sequencing of urinary cells from kidney allograft recipients offers a noninvasive means of diagnosing acute pathologies in the human kidney allograft.

Fig 1. Maximum error for detecting the DNA fraction α in a simulated DNA sequencing experiment by varying sequencing depth and number of SNVs.

Maximum absolute (A) and maximum relative (B) errors are represented. A total of 35 scenarios combining five different numbers of SNVs N = {10, 50, 100, 500, 1,000} and seven depth of coverage M = {10, 50, 100, 500, 1,000, 5,000, 10,000} were simulated (S3 Fig in S1 File). Represented here are maximum error observed in 1,000 simulations for every tested α ranging from 0 to 0.5 in steps of 0.01.

Fig 2. Estimation of DNA fraction (Alpha) in a combination of two healthy DNA sources.

Five scenarios of DNA mixtures and three replicates for each scenario were performed. From left to right: 100% from individual S1 and 0% from individual S2; 0% from individual S1 and 100% from individual 2; 50% from individual 1 and 50% from individual 2; 70% from individual S1 and 30% from individual S2; 90% from individual S1 and 10% from individual S2. The estimated fractions (estimated α) are represented by black dots. The expected fractions when DNA concentration in individual S1 was 19 times lower than DNA concentration in individual S2 are represented by red dots. The expected fractions before correction for DNA concentration are represented by blue dots.

Fig 3. Effect of family relationship and ethnicity on detecting DNA fraction in a combination of two DNA sources.

Each dot represents in A) the maximum absolute error and in B) the maximum relative error for each expected (α) from 0 to 0.5 in steps of 0.01 over 100 pairs of siblings (red), 100 pairs of individuals belonging to the same population (green) and 100 pairs of individuals belonging to different populations (blue). Afr = Africans. Amr = Americans. Eas = East Asians. Eur = Europeans. Sas = South Asians.

Fig 4. Donor/recipient to total DNA fraction in urine from 32 real kidney allograft recipients.

Box plots and individual data points of the estimated fraction (observed α) are estimated from deep DNA targeted sequencing of urinary cells. AR: Acute Rejection. ATI: Acute Tubular Injury. A statistically significant difference was observed between all the diagnostic categories (P = 0.035, Kruskal-Wallis test). By Dunn’s test, difference in observed α between the two pathologies and no pathology group was statistically significant: ATI vs no-pathology: P = 0.0064 and AR vs no-pathology: P = 0.026. The pairwise comparison of AR and ATI pathologies was not statistically significant (P > 0.05).