Study Design
This is a pragmatic (13) observational field study designed to model the effects of ischemia and body-cooling times on the viability and function of HSPCs recovered from the BM of deceased organ donors. The study’s external validity (generalizability) was enhanced by securing the participation of multiple OPOs operating under normal field conditions. Except for special training related to the details of bone recovery and shipment (see below), usual procurement conditions were in effect. Because the OPOs were geographically dispersed, the collected data cover the full spectrum of ischemia times likely to be seen under “real-world” procurement and shipping scenarios.
Donor Tissue Procurement And Transport
Previously developed clinical recovery methods combined with subsequent experience in the ongoing VCA transplant immunomodulation clinical trial at Johns Hopkins University (ClinicalTrials.gov Identifier: NCT01459107) formed the basis for the procurement and transport protocols (4, 14–16). However, these protocols required optimization and validation to ensure that multiple OPOs could reliably operationalize them in a manner that allowed for the production of consistent yields of functionally viable HSPCs after cross-country transport of recovered bones. To that end, a streamlined OPO recovery procedure, combined with dedicated kits and centralized training on recovery and shipment procedures were employed.
Recovered bones were shipped to one of two processing facilities located in Centennial, CO (Facility A) or Indianapolis, IN (Facility B). Vertebral sections from T-8 through L-5 (Facility A and B) and/or ilia (Facility A, only) were procured by six OPOs: Gift of Hope (Itasca, IL); Donor Alliance (Denver, CO); Iowa Donor Network (North Liberty, IA); Mid America Transplant (St. Louis, MO); and Nevada Donor Network (Las Vegas, NV). Bones were recovered by OPO personnel using an osteotome and mallet under an IRB approved protocol from research-consented organ and tissue donors. Unprocessed bones were wrapped in lap sponges and towels soaked in saline and placed in triple-sealed bags to ensure moisture retention during shipment. Wrapped specimens were shipped overnight on wet ice.
Manual Debriding
Upon receipt, in an ISO 5 clean room (Facility A) or a Biological Safety Cabinet (Facility B), soft tissue was manually debrided using scalpels and gouges. Once visible, the pedicles were removed using either a tissue processing band saw or a Stryker System 6 Saw (Stryker, Kalamazoo, MI) leaving only the connected vertebral bodies. Using a boning knife (Facility B) or tissue processing band saw (Facility A), vertebral bodies were separated at the intervertebral disc. Remaining intervertebral disc and soft tissue were removed with a scalpel, leaving clean, separated VBs. Ilium soft tissue was removed with gouges and a scalpel. Care was taken to ensure that the cortical bone was not breached to preserve and protect the hypoxic cancellous BM throughout the entire debriding process (Fig. 1).
Using a saw and/or anvil shears, VBs and IL were cut into 5 cm3 pieces small enough for fragmenting with a bone grinder. The pieces were immediately submerged into 500 mL processing medium (Iscove’s Modified Dulbecco’s Medium containing 100 U/mL DNase, 10 U/mL heparin, and 2.5% human serum albumin). IMDM is suitable for rapidly proliferating high-density cell cultures and ideal for supporting T- and B-lymphocytes. DNase is essential for the mitigation of cell clumping as a result of DNA release from dying cells and post-mortem stress on deceased-donor derived BM. Heparin was used as an anticoagulant. HSA provided a protein source to prevent cell adherence and adsorption to surfaces (Fig. 1).
Grinding and Elution
An electric bone grinder was assembled in an ISO-5 cleanroom (Facility A), and a purpose-built bone grinder (Biorep Technologies Inc., Miami, FL) was assembled in a Biological Safety Cabinet (Facility B). In both facilities, a 2 L stainless steel beaker containing 100 mL of fresh processing medium was placed under the grinding head to catch bone fragments and media flow-through. Bone types were kept separate if both VB and IL from the same donor were processed. Processing medium was used to rinse the grinder throughout the process to prevent bone from drying and sticking to the chamber. Once all bone pieces were ground, the chamber was thoroughly rinsed with fresh processing media. The final volume in the stainless-steel beaker was typically around 750 mL
Stainless steel sieves were stacked with a No. 40 (425 µm) on top of a No. 80 (177 µm) and seated over a round catch-pan (WS Tyler, St. Catherines, ON). The stainless-steel beaker was swirled and poured over the sieves. Bone fragments were distributed evenly on top of the sieve and rinsed with 250 mL of fresh processing medium. The sieved BM product, approximately 1000 mL, was transferred to a sterile pack for final analysis.
Nucleated Cell Counts
An aliquot of BM extract was subjected to red blood cell lysis with ammonium chloride RBC lysis buffer. In a 15 mL conical tube, 4 mL of 9% ammonium chloride was added to 1 mL of BM cell suspension and incubated for 5 minutes at room temperature. Following incubation, the lysed sample was filled to the top of the tube with IMDM containing 100 U/mL DNase, 10 U/mL heparin, and 2.5% HSA processing medium. The lysed sample was centrifuged at 300 x g for 5 minutes and decanted. The sample was then washed with 15 mL of processing medium, centrifuged at 300 x g for 5 minutes, and decanted. Finally, the lysed cells were re-suspended with 1 mL of the same processing medium. Viable nucleated cell counts were obtained using Trypan blue and a hemocytometer.
Flow Cytometry
Flow Cytometry was performed using an ACEA Biosciences NovoCyte 2060R equipped with 488 nm and 640 nm lasers. ISHAGE methods were used to enumerate CD45 + and CD34 + cells (16). 500 µL of lysed bone marrow extract was stained for 15 minutes with 2 µL each of CD45-FITC, CD34-APC, 7-AAD, and Annexin-PE. All conjugated antibodies were purchased from BD Biosciences and 7-AAD was obtained from Tonbo Biosciences. Cells were also stained with individual conjugate antibodies for controls and compensation. After incubating for 15 minutes, cells were washed with Dulbecco’s phosphate buffered saline, centrifuged, and re-suspended in 500 µL of PBS. These samples were run directly on the flow cytometer and analyzed using the ISHAGE gating scheme (16) For each sample 100,000 total events gated on the Singlets gate were collected.
Colony Forming Unit (CFU) Assay
The concentration of RBC lysed cell suspension was first adjusted to 105 viable cells/mL with processing medium before adding 250 µL to 2.5 mL of semisolid medium, Methocult Optimum (Stem Cell Technologies, Vancouver, Canada) and then vigorously vortexed to achieve adequate mixing. A 3 cc syringe was used to remove at least 2.2 mL of Methocult containing cells. 1.1 mL was dispensed into each of two 35 mm non-tissue culture treated dishes. The dishes were covered and tilted to ensure coating of the entire plate surface with Methocult. The two dishes were placed inside a larger 100 mm petri dish with a third uncovered 35 mm dish containing sterile deionized water to humidify the plate. Plates were incubated for 14 days at 37oC, 5% CO2 before scoring colonies.
Numbers of donors and bone marrow samples utilized for statistical modeling
Seventy-five bones from 62 donors were initially received at one of the two BM processing facilities. The numbers of samples with complete data records differed depending on the outcome being modeled. Table 1 provides a breakdown of the numbers received and the numbers with complete data available for statistical modeling by outcome.
Table 1
Numbers of donors and bones with complete data records available for analysis
| Modeled Outcome | Donors | Bones | VB | IL |
| %CD34+ | 62 | 75 | 52 | 23 |
| CFU-TOTAL/105 | 54 | 67 | 42 | 25 |
| CFU-GM 105 | 54 | 66 | 41 | 25 |
Definition Of Ischemia Time
Total ischemia was defined as the interval from time of death (when the donor’s arterial system was cross-clamped and circulation ceased) to start of BM recovery at the processing facility. For purposes of statistical modeling, this total interval was separated into three successive and mutually exclusive time components: (a) Warm-Ischemia Time (WIT): Beginning at time of death and ending either when bones were recovered and packed on ice, or when the body was placed in a cooler. (b) Body-Cooling Time (BCT): Beginning when the body was placed in the cooler and ending when recovered bones were packed on ice. (c) Cold-Ischemia Time (CIT): Beginning when recovered bones were packed on ice and ending when processing began for extraction of HSPCs. Thus, Total Ischemia Time = (WIT) + (BCT) + (CIT). When body cooling was not used, BCT was coded zero and Total Ischemia Time = (WIT) + (CIT). Ischemia times were considered the main variables of interest in statistical models.
Definition Of Experience
Because this was the first series in our hands in which BM was processed from the bones of deceased donors, we hypothesized that HSPC quality might improve with learning as we gained more processing experience. This hypothesis rests on long-established research demonstrating that learning curves exert significant effects on outcomes and costs in both industrial manufacturing (17) and medical practice settings (18–20). To control for learning, we created a variable, EXPERIENCE, defined as the number of donors processed prior to the current one. For the ith donor, EXPERIENCE was coded i – 1, to indicate that EXPERIENCE is always one less than the serial number of the current case being processed. Because Facility A began processing BM five months before Facility B, and because Facility B had the advantage of participating in and learning from cases processed at Facility A, we hypothesized that the two facilities would have different learning trajectories. To account for this possible difference, each facility’s experience was coded separately. To identify the facilities in the model, we coded Facility A = 1 and Facility B = 0. The effect of EXPERIENCE was initially estimated in separate regression models and subsequently incorporated as a covariate in final adjusted models to control for the effect of learning on outcomes (see Online Supplement, Technical Appendix A for details).
Other Covariates
Other variables tested in statistical models were: (1) BONE TYPE, vertebral bodies (VB) and ilia (IL), (representing the two sources of BM cells, coded VB = 1; IL = 0); DONOR SEX (percent male); and DONER AGE (years). These additional covariates were treated as exogenous factors and were included in final models only if they were statistically significant, or they improved the model’s performance.
Outcome Variables
Outcomes were defined according to three hallmarks of potential in vivo utility: (a) The proportion of recovered CD34 + cells that were viable (%CD34+) determined by 7-AAD, (b) The total number of colony forming units (CFUs) per 105 total nucleated cells (TNC) plated (CFU-TOTAL), and (c) The number of CFU granulocyte-macrophages detected per 105 nucleated cells (CFU-GM).
Summary Statistics
Donor and processing-facility characteristics, ischemia times, and outcome measures were summarized as means or percentages as appropriate. Crude (unadjusted) comparisons between FACILITIES (A vs. B), BONE TYPE (VB vs. IL), and BODY COOLING (Yes or No) were tested using independent-groups t-Tests or z-tests for proportions.
Statistical Modeling
The associations of ischemia times with outcomes were initially investigated in unadjusted regression models using only ischemia times as predictors. Additional models were then estimated to determine the separate associations of EXPERIENCE with outcomes. Finally, the effects of ischemia were evaluated in multivariable models that controlled for the potential influences of FACILITY, EXPERIENCE, BONE TYPE, DONOR SEX, and DONOR AGE. Separate models were estimated for each of the three outcomes of interest (%CD34+, CFU-TOTAL, and CFU-GM).
Ordinary least-squares (OLS) linear regression was employed to test a range of candidate models, including models incorporating two-way interactions, as well as logarithmic and second-order polynomial terms. From these candidates, the best reduced models were selected based on the following criteria. (a) Models with the greatest explanatory power (highest R2 values) were favored. (b) Parsimonious models that explained the greatest percentage of variation with the fewest predictors were favored. The adjusted R2, which guards against over-specification by penalizing models containing greater numbers of predictors (21), was used as a comparative indicator of explanatory power in selecting the most parsimonious models. Models that achieved the highest R2 values, while simultaneously maintaining or increasing the adjusted R2, were favored. (c) Models with greater precision, as indicated by relatively smaller standard errors associated with both the model and model coefficients were favored. (d) Models with the best fit, as judged by an assessment of residual plots, were favored. Residuals were plotted and examined visually for discernable patterns, and confirmed quantitatively by regressing residuals onto observed values to uncover possible interactions or underlying curvilinear relationships.
%CD34 + was measured as a proportion, limited to the closed unit interval, [0 ≤ (%CD34+) ≤ 1], which caused traditional OLS linear regression to produce unrealistic fitted values exceeding these interval boundaries. To correct for this, we substituted beta regression (22) for linear regression in models of %CD34+. Beta regression is useful in situations where the response variable is a rate or proportion measured on a continuous scale and bounded by minimum and maximum values. We modeled a transformed variable, pCD34* = [100 x (%CD34+) + 1]/102, which satisfies the distributional assumption of beta regression that the outcome variable must be restricted to the open interval, [0 < (%CD34+) < 1]. So that predicted values could be reported in their original percentage units, beta regression results were back transformed to obtain:
Pred(%CD34+) = [102 x (Pred(pCD34*)) – 1 ∕ 100]
(A technical description of beta regression is provided in the Online Supplement, Technical Appendix B).
Model Validation
All models were validated using leave-one-out bootstrap cross-validation (23), accomplished by randomly omitting one observation with replacement from the dataset and re-estimating the model from the remaining observations. The resulting model was then used to predict the omitted observation. This procedure was repeated 200 times, yielding 200 models with predicted values, model coefficients, standard errors, and 95% confidence intervals. Model parameters were summarized as averages of the 200 bootstrapped models. Since bootstrap models are naïve to the omitted observations, this form of validation serves as an estimate of the predictive accuracy likely to be seen when the original model is used to predict new observations (21). Model coefficients are reported for the original models and compared with averaged coefficients ± 95% confidence intervals from the 200 cross-validated models.