Context.—

The National Institutes of Health Genotype-Tissue Expression (GTEx) project was developed to elucidate how genetic variation influences gene expression in multiple normal tissues procured from postmortem donors.

Objective.—

To provide critical insight into a biospecimen’s suitability for subsequent analysis, each biospecimen underwent quality assessment measures that included evaluation for underlying disease and potential effects introduced by preanalytic factors.

Design.—

Electronic images of each tissue collected from nearly 1000 postmortem donors were evaluated by board-certified pathologists for the extent of autolysis, tissue purity, and the type and abundance of any extraneous tissue. Tissue-specific differences in the severity of autolysis and RNA integrity were evaluated, as were potential relationships between these markers and the duration of postmortem interval and rapidity of death.

Results.—

Tissue-specific challenges in the procurement and preservation of the nearly 30 000 tissue specimens collected during the GTEx project are summarized. Differences in the degree of autolysis and RNA integrity number were observed among the 40 tissue types evaluated, and tissue-specific susceptibilities to the duration of postmortem interval and rapidity of death were observed.

Conclusions.—

Ninety-five percent of tissues were of sufficient quality to support RNA sequencing analysis. Biospecimens, annotated whole slide images, de-identified clinical data, and genomic data generated for GTEx represent a high-quality and comprehensive resource for the scientific community that has contributed to its use in approximately 1695 articles. Biospecimens and data collected under the GTEx project are available via the GTEx portal and authorized access to the Database of Genotypes and Phenotypes; procedures and whole slide images are available from the National Cancer Institute.

The Genotype-Tissue Expression (GTEx) project launched in 2010 as a National Institutes of Health (NIH) Common Fund study that aimed to (1) create a reference of gene expression across “normal” human tissues and (2) elucidate how genetic variation influences gene expression. GTEx is led by the National Human Genome Research Institute, the National Institute of Mental Health, and the National Cancer Institute. The scientific objectives and operational details of GTEx have been described previously.1–4  The project has established a public atlas of human gene expression and its regulation across multiple tissue types, along with an associated biobank of tissues, cell lines, and nucleic acid derivatives for the scientific community. All protected-access data, including genomic profiling and clinical data for each eligible donor, are available through the Database of Genotypes and Phenotypes (dbGaP) (http://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000424). All open-access derived and analyzed data are publicly available on the GTEx portal (www.gtexportal.org). This rich dataset has enabled the research community to analyze the effects of genetic variation and gene expression on a tissue type–specific basis and has aided in the functional interpretation of genetic associations with disease. Since the initial RNA sequencing of more than 1000 samples from the GTEx pilot phase,2  2057 data access requests to dbGap from established researchers have been authorized, and 1695 manuscripts utilizing GTEx data have been published and indexed in PubMed.

The GTEx project included contributions from numerous groups with diverse expertise that collectively comprise The GTEx Consortium. GTEx applies a comprehensive approach to biospecimen collection and processing, pathology review, molecular analysis, and data management. A schematic representation of the biospecimen and data workflow for the GTEx project can be found in Supplemental Figure 1 (see supplemental digital content at https://meridian.allenpress.com/aplm in the March 2025 table of contents). Biospecimens were procured at several medical centers and organ procurement organizations (collectively referred to as biospecimen source sites [BSSs]).2,3  A central biorepository at the Van Andel Institute created and shipped collection kits and received, processed, and shipped biospecimens to the analysis laboratory. All BSSs used the same standard operating procedures (SOPs),5,6  and a single laboratory (Broad Institute of Massachusetts Institute of Technology [MIT] and Harvard, Cambridge, Massachusetts) performed molecular analysis under controlled SOPs, thereby minimizing the variability associated with different workflows and equipment for preservation, nucleic acid isolation, whole-exome and whole-genome sequencing, and messenger RNA (mRNA) and small RNA sequencing. Quality control of data obtained from GTEx specimens was rigorous and included verification of donor and tissue type. RNA sequencing data from tissue samples underwent both sex incompatibility checks and nucleic acid fingerprinting to confirm samples were from the correct donor. Additionally, within-tissue sample-to-sample correlations of expression levels were computed and compared with histopathologic sample assignment to confirm that the correct tissue had been sampled and analyzed.2 

The standardized biospecimen collection and analysis practices applied during the GTEx project served to minimize preanalytic variability associated with specimen-related factors and their potential impact on analytic endpoints. However, when postmortem tissue is used for research, differences among donors are essentially unavoidable. Postmortem donors differ in the cause and rapidity of death, and delays from death to biospecimen preservation are difficult to control. Available evidence suggests that postmortem interval (PMI), the duration of time between death and tissue preservation, may adversely affect DNA integrity,7  RNA integrity,7–9  mRNA levels,8  micro RNA levels,10  pH in brain tissue,11  and ultrastructural morphologic details that are indicative of autolysis severity (cellular destruction by intracellular enzymes).12,13  However, the timing and magnitude of PMI-associated effects remain unclear given that other studies have reported an absence of effect14–17  or tissue-specific effects.18–20  A comprehensive picture of PMI effects has proved elusive as individual studies differ from one another in the tissues examined, sample sizes, and PMI ranges investigated. Conversely, potential effects of the rapidity of death (also called the agonal phase) have received comparatively less attention. Although many studies focus solely on brain specimens, it remains unclear whether autolysis and/or RNA integrity are12,19  or are not21  affected by the rapidity of death of the donor. To explore potential relationships between the agonal phase and RNA integrity or the severity of autolysis, a modified scoring system developed by Hardy et al22  was used to record the rapidity of death for GTEx donors.

Each of the nearly 30 000 tissue specimens collected during the GTEx project was reviewed by a team of American Board of Pathology–certified pathologists to ensure samples were representative of the target tissue, free of disease, and of suitable quality for gene expression analysis. This histologic review revealed several unexpected lesions in grossly normal-appearing tissue that are discussed in a companion article.23  In this report, we summarize the challenges encountered during tissue procurement and strategies applied that proved beneficial during this large multicenter study. Further, we utilized this data set to evaluate the effects of tissue type and preanalytic factors, including PMI and rapidity of death, on autolysis and RNA integrity markers. Examination of quality metrics within the context of postmortem preanalytic factors revealed tissue-specific differences in autolysis and RNA integrity markers and susceptibilities to PMI and rapidity of death.

Study Inclusion and Exclusion Criteria

For inclusion in the GTEx project, next-of-kin or the donor’s legally authorized representative provided research authorization after death of the donor.24  The criteria used to assess donor eligibility has been previously described1,25 ; however, ultimately the donor inclusion/exclusion criteria applied were specific to each GTEx analytical study based upon its aims and analytic approach. In the present study, donors were between 21 and 70 years of age and had a body mass index between 18.5 and 35. Any of the following resulted in exclusion from the study: a diagnosis of metastatic cancer; treatment with chemotherapy or radiation for cancer or any other condition within 2 years of death; a blood transfusion within 48 hours of death; active sepsis; exposure to human immunodeficiency virus/acquired immunodeficiency syndrome (HIV/AIDS), hepatitis C virus, or hepatitis B virus within the 5 years preceding death, including intravenous drug use, or repeated positive reactive screening tests for HIV-1 or HIV-2 antibodies; an inconsistency between documented and genetically confirmed sex; and a documented or genetically confirmed severe congenital pathologic disorder or condition. Individual tissue samples were considered unacceptable and excluded from analysis if they were not representative of the target tissue (ie, were of an insufficient amount and/or included a high proportion of extraneous tissue), displayed evidence of disease, or were from a transplanted organ. Donors were classified as either a brain donor or nonbrain donor depending on whether brain tissue could be collected; brain tissue was not collected from donors who were on a ventilator for ≥24 hours.26 

Tissue Procurement and Preservation

GTEx prosectors were experienced staff at their respective BSSs who underwent training and followed detailed guidance outlined in GTEx SOPs,5  which specified the preferential collection of normal-appearing tissue and instructions to record any abnormalities observed. Additionally, BSSs provided and received ongoing feedback from project pathologists. In total, samples from up to 40 tissue types were collected from 981 donors (965 postmortem donors and 16 organ donors) at several BSSs across the United States. However, samples from 956 donors were used in the present study, as 25 donors were determined to be ineligible based on the criteria described above. The rapidity of death, or the duration of the agonal phase, was recorded for each donor when possible using the Hardy scale, with the addition of a score “0” that indicates mechanical ventilation was received prior to death (Table 1; range = 0–4).22  The exact number of biospecimens procured per donor varied due to organ donation, specificity of tissues consented for research, and manner of death. The total numbers of acceptable samples by tissue type that were collected under the GTEx project from eligible donors are summarized by sex in Figure 1. PMI was recorded as the number of minutes that elapsed from the time of death or withdrawal from life support to immersion of the tissue specimen in PAXgene fixative solution and was recorded for individual tissues.

Figure 1.

The number of acceptable tissue specimens collected during the Genotype-Tissue Expression project from eligible male and female donors by tissue type. The combined bar height reflects the total number of specimens of each tissue type. In total, 40 different tissue types were collected and preserved in PAXgene fixative, although 6 tissue types were specific to females and 2 were male-specific. While the number of tissues collected per donor varied due to difficulties encountered during collection, discontinuation of several tissue types, and designation of brain and nonbrain donors, on average 26 tissues were collected per donor.

Figure 1.

The number of acceptable tissue specimens collected during the Genotype-Tissue Expression project from eligible male and female donors by tissue type. The combined bar height reflects the total number of specimens of each tissue type. In total, 40 different tissue types were collected and preserved in PAXgene fixative, although 6 tissue types were specific to females and 2 were male-specific. While the number of tissues collected per donor varied due to difficulties encountered during collection, discontinuation of several tissue types, and designation of brain and nonbrain donors, on average 26 tissues were collected per donor.

Close modal

SOPs for nonbrain donors required that the first tissue be collected within 8 hours of cardiac cessation or the recorded time of death and placed in PAXgene fixative solution (PAXgene Tissue System; Qiagen, Germantown, Maryland). SOPs for brain donors required that all tissues be placed into PAXgene fixative within 24 hours of cardiac cessation.26 

SOPs specified that samples be collected from tissue that appeared grossly normal. If no normal tissue was apparent, the organ was sampled and abnormalities were recorded. In general, the thickness of individual tissue aliquots were ≤4 mm. To optimally maintain tissue morphology and RNA and DNA quality, each GTEx tissue was divided into 2 tissue blocks, 1 for histology and 1 for molecular analysis; both tissue blocks were preserved in PAXgene Tissue Fixative (Qiagen) solution for 6 to 24 hours, followed by PAXgene Tissue Stabilizer (Qiagen) according to project-specific SOPs.26  The PAXgene Tissue System includes a proprietary fixative that provides excellent preservation of both histomorphology (Figure 2, A) and nucleic acids27  while obviating the need for dry ice or liquid nitrogen during shipment of biospecimens from remote BSSs and eliminating concerns of protein and biomolecule crosslinking associated with formalin fixation. For brain donors, cortex (the frontal pole region of the frontal cortex) and cerebellum (cerebellar hemisphere) specimens28  were collected and placed in PAXgene fixative for histologic and quality analysis prior to shipment of the remaining whole brain specimen on dry ice to the University of Miami Brain Endowment Bank for expert dissection, study, and distribution to the Laboratory, Data Analysis, and Coordinating Center (LDACC) for molecular analysis. For each tissue type, 1 PAXgene-preserved tissue block was shipped directly to the LDACC at the Broad Institute of MIT and Harvard University for isolation of nucleic acids and for molecular analysis; upon receipt, PAXgene-preserved tissues were stored at −80°C until characterization. For histology, the remaining PAXgene-preserved tissue block, which was adjacent to blocks utilized for molecular studies, was sent to the Comprehensive Biospecimen Resource (CBR, Van Andel Institute, Grand Rapids, Michigan) and then processed and embedded in paraffin and stored at −20°C to optimally preserve molecular analytes for future studies and pathology review. For some studies outside the scope of this manuscript, additional tissue blocks were flash frozen and shipped to the CBR.

Figure 2.

Representative micrographs of hematoxylin-eosin–stained PAXgene-fixed, paraffin-embedded (PFPE) tissue that was collected during the Genotype-Tissue Expression project. (A) Digital image of stomach, fundus, and mucosa that displayed excellent preservation of glandular components showing foveolar, parietal, and chief cells. (B) Digital image of a pancreas specimen with a slight degree of autolysis (autolysis score = 1). There is uniform separation of the acinar glands but good overall staining and structure of cytoplasm and nuclei. Two well-preserved endocrine islets and a portion of a third, as well as a small duct (upper left), are also present. The corresponding RNA integrity number (RIN) was 6.3. (C) Digital image of a pancreas specimen with severe autolysis (autolysis score = 3). Despite the strong nuclear and cytoplasmic staining, the acini have lost their cohesion, and their cells are largely dissociated. The corresponding RIN was 2.0 (hematoxylin-eosin, original magnification ×20).

Figure 2.

Representative micrographs of hematoxylin-eosin–stained PAXgene-fixed, paraffin-embedded (PFPE) tissue that was collected during the Genotype-Tissue Expression project. (A) Digital image of stomach, fundus, and mucosa that displayed excellent preservation of glandular components showing foveolar, parietal, and chief cells. (B) Digital image of a pancreas specimen with a slight degree of autolysis (autolysis score = 1). There is uniform separation of the acinar glands but good overall staining and structure of cytoplasm and nuclei. Two well-preserved endocrine islets and a portion of a third, as well as a small duct (upper left), are also present. The corresponding RNA integrity number (RIN) was 6.3. (C) Digital image of a pancreas specimen with severe autolysis (autolysis score = 3). Despite the strong nuclear and cytoplasmic staining, the acini have lost their cohesion, and their cells are largely dissociated. The corresponding RIN was 2.0 (hematoxylin-eosin, original magnification ×20).

Close modal

Pathology Review

A hematoxylin-eosin–stained slide was generated at the CBR from the histology tissue block of each tissue sampled from all GTEx donors. Digitally scanned whole slide images of PAXgene-fixed/stabilized, paraffin-embedded tissue sections were created using Aperio Scanscope software (Leica Biosystems). These whole slide digital images and, if necessary, the parent glass slide were reviewed by 1 of 4 project pathologists (M.B., P.A.B., L.S., Maria Tomaszewski, MD) at the Pathology Resource Center that was established for the project. Evaluation focused on (1) verification of the topographic site, (2) presence and degree of autolysis, (3) presence and amount of extraneous tissue, (4) presence of pathologic findings, and (5) the size and number of tissue pieces (Table 2). During histologic evaluation, the degree of autolysis present in the target tissue within the section was quantified by the reviewing pathologist using a scoring system (0 = none, 1 = slight, 2 = moderate, 3 = severe). Figure 2, B and C, contain representative micrographs of slight (score of 1) and severe (score of 3) autolysis, respectively. The histologic review of each biospecimen was paramount as a form of quality control to verify the site of origin, content, quantity, purity, and morphologic integrity of tissue samples. The results of this review were annotated in the Comprehensive Data Resource, a custom database designed for the project, and transmitted electronically to the LDACC, which carried out molecular analysis of adjacent PAXgene-fixed and stabilized tissue aliquots. As an added measure of ongoing quality improvement, procurement feedback was provided to the prosectors at the BSSs (Supplemental Figure 1).

Clinical Data

De-identified clinical data for each donor were captured and included as part of the electronic report for each GTEx case (Supplemental Figure 2 provides an example) and stored in the Comprehensive Data Resource hosted at Leidos Biomedical Research, along with associated annotated digital images, stored initially at the Van Andel Institute but currently available through the GTEx tissue image library (https://brd.nci.nih.gov/brd/image-search/searchhome). Information on clinical diagnoses, cause of death, and Hardy score of each consented qualifying donor was also captured and has been made available by the LDACC through dbGaP under controlled access.

RNA Isolation

At the LDACC, total RNA was isolated from the PAXgene-preserved tissue aliquots designated for molecular analysis and stored at −80°C. Total RNA was isolated from 20 mg of breast and adipose tissue, 10 mg of spleen, 5 mg of pancreas, and 10–12 mg of all other tissue types using Qiagen’s PreAnalytix PAXgene Tissue RNA/miRNA Kit as previously described.2  RNA quantity and purity were assessed using a Nanodrop 8000 spectrophotometer (Thermo Scientific, Waltham, Massachusetts). RNA integrity was quantified as an RNA integrity number (RIN) that was obtained with an Agilent 2100 Bioanalyzer (Santa Clara, California).

Statistical Analysis

Potential associations between PMI, Hardy score, RIN, and autolysis score were investigated by Pearson correlation analysis using R software (version 4.0.2).29  Statistical significance was set at P < .00025 based on the Bonferroni-corrected P value for all tissues and tests compared. The confidence interval was also adjusted accordingly and is reported at a level of 99.975%.

GTEx Donor Characteristics

During the GTEx project, a total of 26 468 tissue samples (yielding more than 55 000 tissue aliquots) were collected from the 956 eligible donors for this study. An average of 26 tissues were collected from each eligible donor; however, the final tally of specimens collected for each tissue type/subtype differed due to sex, tissue-specific challenges in dissection, discontinued collection of several tissue types (kidney medulla, urinary bladder, fallopian tube, endocervix, and ectocervix; Supplemental Table 1), and organ transplant donation (Figure 1).

Donor PMI, represented as the ischemic time of the first tissue collected from a donor, ranged between <1 and 27 hours. Based on preliminary results of the GTEx pilot study,1  an effort was made to collect tissues from the nonbrain donors within 8 hours of death. This resulted in different PMI distributions for brain and nonbrain donors (Figure 3, A): 486 of the 541 (90%) eligible nonbrain donors had a PMI of less than 8 hours compared to just 17 of the 415 (4%) eligible brain donors.

Figure 3.

Donor distribution by postmortem interval (PMI) and modified Hardy score. (A) The number of brain and nonbrain donors categorized by donor PMI, which was defined as the ischemic time (in hours) of the first tissue sample collected from the donor. (B) Distribution of eligible donors by modified Hardy score and donor PMI (<8 hours or ≥8 hours).

Figure 3.

Donor distribution by postmortem interval (PMI) and modified Hardy score. (A) The number of brain and nonbrain donors categorized by donor PMI, which was defined as the ischemic time (in hours) of the first tissue sample collected from the donor. (B) Distribution of eligible donors by modified Hardy score and donor PMI (<8 hours or ≥8 hours).

Close modal

The distribution of eligible GTEx donors based on the modified Hardy score was disproportionate, as 499 of the 939 donors (54%) for which rapidity of death was recorded had a Hardy score of 0, indicating they were on a ventilator immediately prior to death, while only 35 of 939 (4%) donors had a Hardy score of 1 (Figure 3, B). Donors who received mechanical ventilation prior to death (Hardy score = 0) tended to have a shorter donor PMI than those with a Hardy score of 1–4; donors with a PMI <8 hours represented 91% (455 of 499) of individuals with a Hardy score of 0 but just 3%–13% (8 of 236 to 15 of 115) of those with a Hardy score of 1–4 (Figure 3, B). These differences in donor PMI may be attributable to a greater representation of nonbrain donors among individuals with a Hardy score of 0 (Supplemental Figure 3).

Tissue-Specific Differences in RNA Integrity and Autolysis

The integrity of individual GTEx tissue samples was evaluated by assessing RNA integrity (via RIN) and the degree of autolysis (determined by a pathologist and quantified by autolysis score). While the majority of the acceptable tissue samples analyzed displayed a RIN ≥6 (13 913 of 23 206 samples, 60%), indicating RNA was of suitable quality for downstream molecular analysis, tissue-dependent differences in RIN were observed (Figure 4). In total, 24 tissue types had an average RIN that was ≥6, while the remaining 16 tissues had an average RIN that ranged between 4.34 and 5.96. Tissue types with the lowest average RINs included kidney (cortex, 4.34; medulla, 4.73), spleen (4.63), and pancreas (5.20). The highest average RINs were observed in skeletal muscle (7.70), minor salivary gland (7.48), and esophagus mucosa (7.18). The median was 6.23, which corresponded with the average RIN for tibial nerve tissue.

Figure 4.

Average RNA integrity number (RIN) and average autolysis score of each type of tissue collected from eligible donors (n = 956) in the Genotype-Tissue Expression project.

Figure 4.

Average RNA integrity number (RIN) and average autolysis score of each type of tissue collected from eligible donors (n = 956) in the Genotype-Tissue Expression project.

Close modal

To elucidate tissue-specific differences in tissue quality, the degree of autolysis was assessed in GTEx tissue that appeared healthy during gross examination. The severity of autolysis present varied considerably among the 40 tissue types examined, with autolysis prominent in some (eg, pancreas and gastrointestinal mucosa) and much less evident in others (eg, nerve, skin, adipose tissue, and skeletal muscle) (Figure 4). Interestingly, tissue types that had high RIN values, such as skeletal muscle, tended to have a low degree of autolysis, while those with lower RIN values tended to have a high average autolysis score. The same was true when all tissue types were pooled for analysis, as average RIN values declined with more severe autolysis (Figure 5, A). A potential relationship between the 2 quality markers was explored in 2 tissue types that fall on different ends of the RIN spectrum and display different degrees of autolysis—skeletal muscle (Figure 5, B) and pancreas (Figure 5, C). Results were tissue-specific, as RIN and autolysis score were not correlated in skeletal muscle specimens (r = −0.071; 99.975% CI, −0.19 to 0.048; P = .029; n = 940) but were strongly and negatively correlated in pancreas specimens (r = −0.70; 99.975% CI, −0.76 to −0.63; P < .00001; n = 768) (Figure 5, B and C, respectively). Results suggest that the severity of autolysis may be associated with the degree of RNA degradation in some tissue types, including pancreas.

Figure 5.

Relationship between the quality markers autolysis score and RNA integrity number (RIN) in Genotype-Tissue Expression (GTEx) project tissue samples. (A) Samples from all tissue types were pooled, and the average RIN was calculated for each autolysis score. Average RIN declined with higher autolysis scores in GTEx samples. (B) No correlation was found between autolysis score and RIN among skeletal muscle specimens (n = 967 donors; Pearson correlation coefficient = −0.071; P = .029). (C) Autolysis score and RIN were negatively correlated among pancreas specimens (n = 797 donors; Pearson correlation coefficient = −0.70; P < .00001).

Figure 5.

Relationship between the quality markers autolysis score and RNA integrity number (RIN) in Genotype-Tissue Expression (GTEx) project tissue samples. (A) Samples from all tissue types were pooled, and the average RIN was calculated for each autolysis score. Average RIN declined with higher autolysis scores in GTEx samples. (B) No correlation was found between autolysis score and RIN among skeletal muscle specimens (n = 967 donors; Pearson correlation coefficient = −0.071; P = .029). (C) Autolysis score and RIN were negatively correlated among pancreas specimens (n = 797 donors; Pearson correlation coefficient = −0.70; P < .00001).

Close modal

Effects of PMI on RNA Integrity and Autolysis

Effects of PMI on RNA integrity and autolysis were examined in the collective sample set (samples from 40 tissue types were pooled for analysis) as well as in representative tissue types. When all tissue samples were considered together, longer ischemic times were associated with lower average RINs (Figure 6, A). The average RIN of samples with a PMI <8 hours was more than 6 (6.11–7.22), suggesting RNA from these specimens is more likely to be of suitable quality for RNA sequencing.1,30  Conversely, the average RIN was progressively lower for samples with a PMI ≥8 h (5.33–5.86). Examination of representative tissues suggests that the relationship between PMI and RNA integrity is tissue dependent. Average RINs remained stable in both nerve and skeletal muscle samples across the PMI ranges examined (tibial nerve: 90–1739 minutes; skeletal muscle: 97–2075 minutes) but exhibited a substantial nonlinear decline in pancreas specimens (49–1572 minutes) (Supplemental Figure 4). In pancreas, the largest difference in RIN occurred when samples with an ischemic time of 0–4 hours and 16–20 hours were compared (6.96 versus 3.57), reflecting a decline of approximately 49%. Tissue-specific differences in the proportions of specimens with high (RIN >6), intermediate (RIN 3–6), and low quality (RIN <3) RNA were also observed, as were tissue-dependent changes with progressive ischemia. The percentage of nerve specimens with a RIN >6 displayed a modest decline with PMI, from 70% (90 of 129 specimens) at a PMI <4 hours to 58% (50 of 86 specimens) at a PMI ≥20 hours. (Figure 6, B). On the other hand, the proportion of skeletal muscle specimens with high-quality RNA (RIN >6) remained stable across the PMI range evaluated (86%–95%, 74 of 86 and 245 of 258 specimens, respectively) (Figure 6, C). In pancreas, there was also an incremental shift in the fraction of specimens with high-quality RNA (RIN >6) to those with intermediate (RIN 3–6) and low-quality RNA (RIN <3) observed with longer PMIs. The percentage of pancreas specimens with high-quality RNA (RIN >6) greatly decreased when PMI was ≥8 hours (17%, 75 of 447 specimens) compared to shorter PMIs (83%, 263 of 318 specimens) (Figure 6, D). Statistical analysis supported tissue-dependent effects of PMI on RNA quality. While a weak negative correlation was observed in skeletal muscle specimens (r = −0.13; 99.975% CI, −0.24 to −0.0066; P = .00011; n = 940), PMI and RIN were modestly and negatively correlated in pancreas (r = −0.68; 99.975% CI, −0.75 to −0.60; P < .00001; n = 768) (Figure 6, E and F).

Figure 6.

Postmortem interval (PMI) effects on RNA integrity number (RIN) in Genotype-Tissue Expression tissue samples. (A) Samples from all tissue types were pooled, and the average RIN was calculated for each PMI binned range (displayed in hours). Average RIN was lower among samples with a PMI ≥8 hours. The fraction of samples that yielded RNA of high (RIN >6), moderate (RIN 3–6), and low (RIN <3) quality in (B) tibial nerve, (C) skeletal muscle, and (D) pancreas tissue specimens. (E) RIN and PMI were weakly and negatively correlated in skeletal muscle specimens (n = 967 donors; Pearson correlation coefficient = −0.13; P = .00011). (F) RIN and PMI were negatively correlated in pancreas specimens (n = 797; Pearson correlation coefficient = −0.68; P < .00001). See Supplemental Table 5 for sample sizes.

Figure 6.

Postmortem interval (PMI) effects on RNA integrity number (RIN) in Genotype-Tissue Expression tissue samples. (A) Samples from all tissue types were pooled, and the average RIN was calculated for each PMI binned range (displayed in hours). Average RIN was lower among samples with a PMI ≥8 hours. The fraction of samples that yielded RNA of high (RIN >6), moderate (RIN 3–6), and low (RIN <3) quality in (B) tibial nerve, (C) skeletal muscle, and (D) pancreas tissue specimens. (E) RIN and PMI were weakly and negatively correlated in skeletal muscle specimens (n = 967 donors; Pearson correlation coefficient = −0.13; P = .00011). (F) RIN and PMI were negatively correlated in pancreas specimens (n = 797; Pearson correlation coefficient = −0.68; P < .00001). See Supplemental Table 5 for sample sizes.

Close modal

Given that the onset of autolysis begins upon death, a potential relationship between PMI and the severity of autolysis was examined in grossly normal-appearing tissues. In the collective sample set (samples from 40 tissue types were pooled), average autolysis scores were higher, indicating more severe autolysis, in specimens with longer PMIs (1.39 versus 0.78, PMI ≥20 hours and PMI <4 hours, respectively) (Supplemental Figure 5). When representative tissue types were examined, autolysis scores in skeletal muscle and nerve were relatively stable for the PMI ranges evaluated, while pancreas displayed a progressive increase in autolysis score with longer PMIs (Figure 7, A). The average autolysis score for pancreas specimens increased approximately 2.5-fold for the ischemic range examined (from 0.96 for a PMI <4 hours to 2.63 for a PMI ≥20 hours). PMI affected the distribution of specimens with no, mild, moderate, and severe autolysis in a tissue-specific manner. In tibial nerve specimens, modest changes in the proportion of specimens with no (42%–60%, 36 of 86 and 155 of 258 specimens), mild (40%–57%, 102 of 258 and 49 of 86 specimens), and moderate (0.5%–2%, 1 of 258 and 3 of 166 specimens) autolysis occurred across the PMI range examined (Figure 7, B); notably, severe autolysis was not present in any of the nerve specimens examined regardless of PMI. In skeletal muscle, while the proportion of specimens with no, mild, and moderate autolysis remained stable across the PMI ranges examined, a subtle decrease in the percentage of specimens with no autolysis was observed when those with a PMI <8 hours (57%, 225 of 398 specimens) were compared to those with a PMI ≥8 hours (47%, 251 of 531 specimens) (Figure 7, C). The occurrence of moderate autolysis was first observed among skeletal muscle specimens with a PMI ≥8 hours. In pancreas, specimens with longer ischemic times had a larger fraction of specimens with severe autolysis than those with shorter durations, representing 1% (3 of 206) of specimens with a PMI <4 hours compared to 56% (48 of 85) of those with a PMI ≥20 hours (Figure 7, D). Temporally, declines in the proportion of pancreas specimens with mild and moderate autolysis were first noticeable when specimens with PMIs <8 hours and ≥8 hours were compared: 86% (273 of 318) versus 17% (81 of 486) (mild autolysis) and 8% (25 of 318) versus 52% (254 of 486) (moderate autolysis) of pancreas specimens, respectively (Figure 7, D). A potential correlation between PMI and autolysis score was investigated in skeletal muscle and pancreas specimens. While no correlation was present among skeletal muscle specimens (r = 0.094; 99.975% CI, −0.025 to 0.21; P = .0037; n = 944) (Figure 7, E), autolysis score and PMI exhibited a strong positive correlation in pancreas (r = 0.72; 99.975% CI, 0.65 − 0.78; P < .00001; n = 807) (Figure 7, F). Collectively, results support that the susceptibility of autolysis severity to ischemic time is tissue specific.

Figure 7.

Postmortem interval (PMI) effects on autolysis in representative tissue types. (A) Average autolysis score was calculated for representative tissue types (skeletal muscle, tibial nerve, pancreas) based on the ischemic time of individual tissue samples (PMI, displayed in hours). The fraction of (B) tibial nerve, (C) skeletal muscle, and (D) pancreas samples based on autolysis severity (0 = no autolysis, 1 = slight autolysis, 2 = moderate autolysis, 3 = severe autolysis). (E) Autolysis score and PMI were not correlated in skeletal muscle specimens (n = 966 donors; Pearson correlation coefficient = 0.094; P = .0037). (F) Autolysis score and PMI were positively correlated in pancreas specimens (n = 797; Pearson correlation coefficient = 0.72; P < .00001). See Supplemental Tables 5 and 6 for sample sizes.

Figure 7.

Postmortem interval (PMI) effects on autolysis in representative tissue types. (A) Average autolysis score was calculated for representative tissue types (skeletal muscle, tibial nerve, pancreas) based on the ischemic time of individual tissue samples (PMI, displayed in hours). The fraction of (B) tibial nerve, (C) skeletal muscle, and (D) pancreas samples based on autolysis severity (0 = no autolysis, 1 = slight autolysis, 2 = moderate autolysis, 3 = severe autolysis). (E) Autolysis score and PMI were not correlated in skeletal muscle specimens (n = 966 donors; Pearson correlation coefficient = 0.094; P = .0037). (F) Autolysis score and PMI were positively correlated in pancreas specimens (n = 797; Pearson correlation coefficient = 0.72; P < .00001). See Supplemental Tables 5 and 6 for sample sizes.

Close modal

Effects of Hardy Score on RNA and Tissue Integrity

Potential relationships between Hardy score and tissue quality markers (RIN and the degree of autolysis) were examined collectively (all tissue types), as well as across individual tissue types.

Overall, differences in average RIN and autolysis score among Hardy scores were subtle when all tissues were pooled. While RNA from donors who received mechanical ventilation prior to death (Hardy score of 0) had the highest average RIN (6.78) and the lowest average autolysis score (0.88), relatively subtle differences were observed when Hardy scores of 1–4 were compared (RIN: Δ0.4–0.61; autolysis score: Δ0.03–0.11) (Figure 8, A).

Figure 8.

Effects of Hardy score on RNA integrity number (RIN) and autolysis score in Genotype-Tissue Expression tissue samples. (A) Samples from all tissue types with a recorded donor Hardy score were pooled, and average RIN and autolysis score were calculated for each Hardy score. Average (B) RIN and (C) autolysis scores were calculated for representative tissue types (skeletal muscle, tibial nerve, and pancreas specimens). See Supplemental Tables 2 and 3, respectively, for sample sizes.

Figure 8.

Effects of Hardy score on RNA integrity number (RIN) and autolysis score in Genotype-Tissue Expression tissue samples. (A) Samples from all tissue types with a recorded donor Hardy score were pooled, and average RIN and autolysis score were calculated for each Hardy score. Average (B) RIN and (C) autolysis scores were calculated for representative tissue types (skeletal muscle, tibial nerve, and pancreas specimens). See Supplemental Tables 2 and 3, respectively, for sample sizes.

Close modal

When tissue types were examined individually, average RINs were comparable across Hardy scores for tibial nerve (5.99–6.50) and skeletal muscle (7.24–8.00) specimens, whereas pancreas specimens collected from donors who received mechanical ventilation prior to death had a higher average RIN (6.40) than those from donors with a Hardy score of 1–4 (3.53–3.92) (Figure 8, B). Similarly, average autolysis scores displayed little variation across Hardy scores among nerve (0.31–0.63) and skeletal muscle specimens (0.34–0.65), but average autolysis scores were lower for pancreas specimens collected from donors with a Hardy score of 0 (1.17) compared to donors with a Hardy score of 1–4 (2.09–2.38) (Figure 8, C).

Potential correlations between Hardy score and RIN and Hardy score and autolysis score were investigated in all 40 tissue types collected from donors with a Hardy score between 1 and 4. Donors with a Hardy score of 0 were excluded from statistical comparisons as a score of 0 was (1) outside of the continuum of the agonal phase and (2) displayed a shorter average PMI (Figure 3, B). Weak, but significant, negative correlations between Hardy score and RIN were observed for the left heart ventricle (r = −0.32; 99.975% CI, −0.48 to −0.15), testis (r = −0.29; 99.975% CI, −0.47 to −0.085]), adrenal gland (r = −0.25; 99.975% CI, −0.42 to −0.069), liver (r = −0.24; 99.975% CI, −0.40 to −0.058), skeletal muscle (r = −0.20; 99.975% CI, −0.36 to −0.022), and pituitary (r = −0.20; 99.975% CI, −0.37 to −0.014) (P < .00025 for all) (Supplemental Table 2). The proportion of specimens with RNA of low, intermediate, and high quality did not vary by Hardy score for pancreas, skeletal muscle, or tibial nerve specimens (Supplemental Figure 6).

When a potential correlation between Hardy score and autolysis score was examined in the 40 tissue types collected from GTEx donors with a Hardy score of 1 to 4, stomach was the only tissue that displayed a weak albeit significant positive correlation (r = 0.27; 99.975% CI, 0.092 − 0.43) (Supplemental Table 3). The proportion of specimens with no, modest, moderate, and severe autolysis did not vary among specimens from donors with a Hardy score between 1 and 4 for skeletal muscle, tibial nerve, or pancreas specimens (Supplemental Figure 7).

Outcome of Whole Slide Digital Imaging and Annotation

The histologic examination and annotation of 25 528 digital tissue images from 980 donors was conducted during a period of 4 years by 4 project pathologists. Brain regions that were frozen during expert dissection did not undergo histologic review, and tissue samples collected from 1 donor were not analyzed due to ineligibility (hepatitis C virus diagnosis). Whole slide digital imaging greatly facilitated review by the pathology team. Benefits of digital imaging included the speed and simplicity of image delivery, eliminating the need for transporting fragile glass slides between sites, and enabling the simultaneous review of micrographs (with or without annotation) online in real time by multiple individuals.

Utilization of digital images streamlined histologic evaluation by the reviewing pathologists, as less time was required for transitions between fields of view, magnifications, and tissue blocks in comparison to the review of conventional glass slides with a microscope. All captured digital images were of high enough quality that separate micrographs for consultation or publication were not required. Further, custom features were developed for the Comprehensive Data Resource that allowed the reviewing pathologist to access and annotate images; quantify specimen size; delineate and quantify tissue regions and subregions, as well as areas of extraneous tissue or lesions; and document autolysis through a single user interface. All findings were recorded by the reviewing pathologist on a single, dual, or split screen per their convenience. A free text field was used to record findings and the rationale for tissue exclusion when merited, as well as to provide constructive feedback to BSS prosectors to improve dissection. Tissue abnormalities were also recorded in the electronic pathology report, which was shared with laboratories performing molecular characterization. Electronic pathology reports were also included in dbGaP entries along with genomic sequence data from each tissue and clinical data about the donor. An example of the content of an electronic report generated by reviewing pathologists at the Pathology Resource Center can be found in the supplemental information (Supplemental Figure 2).

Collectively, these features permitted the collection of more exact morphometric data relative to conventional microscopic examination of glass slides. For example, in Supplemental Figure 8, digital imaging software allowed the reviewing pathologist to delineate the coronary artery from extraneous adipose tissue, which resulted in a more accurate calculation of the area or volume of each respective tissue type. Annotation of the micrographs also allowed delineations to be revisited and shared with other program pathologists. A magnification of ×20 was sufficient for histologic review in most instances. When a higher magnification (×40) was required, glass slides were shipped from the CBR to the Pathology Resource Center for review by a project pathologist due to the prohibitive storage requirements of high magnification digital images (5 GB).

Challenges Encountered During Histologic Review

A program pathologist reviewed each histology tissue block collected under the GTEx project to ensure that the correct tissue was collected, morphologic integrity was assessed, and potential microscopic abnormalities were identified. PAXgene tissue fixation provided excellent histologic detail (Figure 2, A). Vivid hematoxylin-eosin staining easily rivaled that of formalin fixation.

While the aim of the GTEx project was to facilitate study of gene expression in normal tissues, the boundary between normal and abnormal biology can be open to interpretation. Histologic review that revealed microscopic abnormalities commonly found in grossly normal tissues, such as the presence of atherosclerotic plaque; congestion of the lung, liver, or spleen; and hepatic, pancreatic, cardiac, and renal fibrosis, as well as age-related changes that included prostatic hyperplasia, endometrial atrophy, mammary atrophy, and reduced spermatogenesis, did not in and of itself result in the tissue being excluded from additional study. Tissues displaying evidence of minor changes were considered acceptable for molecular analysis, and notes characterizing the abnormalities or variations in tissue structure that were observed microscopically were recorded in the electronic pathology report and included in the clinical data in dbGaP, accompanying the genomic sequence data submitted. Tissue alterations that were considered overwhelming by the team of pathologists resulted in the sample being categorized as unacceptable for molecular analysis; examples of lesions that would merit exclusion from molecular analysis included cancer and extensive cirrhosis, pneumonia, nephrosclerosis, and scarring. Additional details on the tissue abnormalities encountered and their frequency among the GTEx samples are summarized and discussed in a companion article.23 

While PAXgene-fixed, paraffin-embedded tissues collected under the GTEx project were generally of high morphologic quality, histologic review revealed a greater frequency of poor or variable preservation, autolysis, and/or hurdles in procurement among a specific subset of tissue types (Table 3). For example, poor preservation of spleen and renal cortex specimens was frequently encountered and corresponded with the presence of vascular congestion in spleen specimens induced by terminal heart failure or chronic lesions and scarring among kidneys that failed to qualify for organ donation. Preservation was at times inconsistent among lung, stomach, and adrenal tissues, which was attributed to the variable presence of extraneous elements associated with inflammation or hemorrhage, differences in stomach contents, and difficulty in separating the adrenal parenchyma from fat or inadvertent removal of the adrenal gland with the kidney for organ transplant, respectively. Pathologists also observed autolysis more frequently among pancreas and prostate gland specimens. In some instances, tissue-specific characteristics translated into unexpected difficulties in dissection that led to discontinued collection. Following initial analysis of specimens collected during the pilot phase, procurement of 4 tissue types (kidney medulla, bladder, fallopian tube, endocervix, and ectocervix) were discontinued because of a high frequency of excessive autolysis or difficulty in isolating the correct target (Supplemental Table 1).

Of the nearly 30 000 tissues collected from eligible donors under the GTEx program, 5% (1394 of 26 468) were categorized as unacceptable for molecular analysis. Unacceptable tissue samples displayed overwhelming tissue abnormalities indicative of disease (such as extensive cirrhosis or nephrosclerosis), excessive autolysis, or evidence of transplant (10 samples), or were considered to be unrepresentative due to an insufficient amount of target tissue or the presence of extraneous tissue. For example, molecular analysis of a sample that contained extraneous adipose tissue could lead to erroneous results (Supplemental Figure 8). Reasons for unacceptability were not exclusive; in many instances, multiple reasons prompted a sample to be considered unacceptable (Supplemental Table 4). Sampling issues (insufficient target or a high proportion of extraneous tissue) were observed in 83% (1152 of 1394) of unacceptable tissue samples, while excessive autolysis was observed in 28% (393 of 1394), and evidence of disease was observed in 8% (112 of 1394) of unacceptable tissue samples. Notably, the prevalence of these issues depended upon tissue type (Table 4). For example, ileum tissue had the highest proportion of unacceptable samples due to either advanced autolysis or sampling issues such as inclusion of paucity of lymphoid tissue (targeted Peyer patches) or excessive muscularis propria rather than the targeted mucosa. Unacceptable sigmoid colon specimens were often predominantly mucosa as opposed to the targeted muscular propria. Coronary arteries that were primarily comprised of pericardial fat were also categorized as unacceptable. Gastroesophageal junction specimens that lacked mucosa were also deemed unacceptable. The occurrence of excessive autolysis was also disproportionate across tissue types; excessive autolysis was observed in a third of unacceptable sigmoid colon specimens but just 3% (1 of 39) of unacceptable thyroid gland tissues (Table 4). Disease-related rejections were minimal due to the sampling of grossly normal tissue by BSS prosectors, however, incidental findings and significant lesions in GTEx tissue specimens were encountered and are summarized in a companion paper.23 

Tissue specimens collected during the GTEx project and their associated histologic image data represent a well-annotated and diverse collection of normal tissue collectively available to the scientific community. The histologic review of each tissue specimen by an experienced project pathologist was streamlined through the use of digital images that facilitated confirmation of accurate sampling, absence of significant lesions, and the evaluation of tissue integrity, while also revealing challenges in procurement. Although collection and processing were standardized to the extent possible, an acceptable range of Hardy scores and tissue ischemic times permitted an investigation into the potential relationship between these preanalytic factors and RNA and tissue integrity markers. An evaluation of RIN and autolysis severity revealed tissue-specific differences, as well as different susceptibilities to Hardy score and PMI. Incidental histologic findings in GTEx tissue specimens that included age-related changes and unexpected lesions in grossly normal-appearing tissue are addressed in a companion paper.23  Collectively, data derived from GTEx specimens, including results presented here and those of the more than 1695 published studies that include analysis of GTEx specimens and/or data, support the view that postmortem and organ donor–derived tissue can be a rich source of molecular data.

Histologic examination of biospecimens provides an important measure of quality control for their suitability for downstream molecular analysis. Pathologists are in a key position to confirm tissue type and assess specimen purity and preservation. These entrance criteria can save the time and resources that intensive molecular analysis entails and can improve the accuracy of molecular results. Several strategies implemented during GTEx could facilitate pathologic review of specimens in other large or multicenter studies. For example, capturing digital images of whole sections streamlined pathologic review by minimizing the time required to transition between fields of view, magnifications, and slides; by enabling real-time consultation due to online access; and by enabling the collection of more accurate morphometric data, a benefit also reported by others.31–33  Obstacles associated with reliance on digital images were limited and included the initial time required to capture whole slide images, and that, for an endeavor of this size, collection of high-magnification images (×40) was precluded by the large amount of storage and transmission memory required. The GTEx project has provided a valuable hands-on opportunity to evaluate procurement procedures and annotation methods. In particular, it demonstrated that whole slide digital imaging technology can be applied to a large data source with excellent results.

Procuring normal-appearing target tissue for the GTEx project was not without challenge. Although the majority (95%, 25 074 of 26 468) of tissue specimens collected were acceptable for downstream molecular analysis, a small fraction was excluded following histologic review due to sampling errors such as the inclusion of extraneous tissue, insufficient amounts of the target tissue, excessive autolysis, or evidence of significant abnormalities. In many cases, difficulties in procurement centered on a few tissue types; in such instances, the reviewing pathologist’s notes were shared with BSS prosectors in an effort to improve specimen quality. These notes were also recorded in the electronic pathology report of each affected specimen, and these reports are accessible via dbGaP. Transparency relating to the specific challenges encountered during procurement and the essential role histologic evaluation served in securing an accurate sample set are paramount to improving practices to ensure high-quality targeted tissue is available for future studies. The 25 000 digital images of GTEx specimens that are available as view-only micrographs through the GTEx portal (https://gtexportal.org/home/histologyPage), for individual download through the GTEx tissue image library (https://biospecimens.cancer.gov/gtexbiobank/histology_viewer.asp), or for bulk download upon request (email: [email protected]) also represent a valuable publicly available resource for research. A recent study used GTEx data to identify an association between histologic features and genetic variants in thyroid and colon tissues,34  while another identified the cause of unexpected variation in gene expression that was rooted in altered cell ratios of the lung specimens evaluated.35 

Postmortem donation represents the primary source of physiologic normal tissue specimens, and is therefore essential to genomic, epigenetic, and gene expression research in tissue that is free from pathology. The RNA integrity of GTEx specimens, as measured by RIN, was tissue-specific, consistent with reports of other smaller scale studies that investigated RIN in multiple tissue types.7,18,36  The mean RIN of GTEx PAXgene-fixed specimens ranged between 4.34 (cortex of kidney) and 7.70 (skeletal muscle), which fell within the range of RINs reported by others for frozen kidney (4.4–4.63)7,19  and skeletal muscle (4.4–9.01)7,36 ; notably, the overlap in RINs reported for GTEx and other studies despite a difference in preservation methods indicates PAXgene is a suitable alternative to snap-freezing for multi-target analyses. The percentage of specimens with high-quality RNA (mean RIN ≥6–6.5) was lower in this study than previously reported for normal adjacent surgical specimens (60% versus 81.6%).37  This is likely attributable to the use of postmortem tissues, as reduced RINs and transcript levels have been observed in postmortem compared to antemortem specimens.38 

PMI is a preanalytic factor irrevocably linked to research tissue that has been collected from a postmortem donor; however, studies are in conflict on the timing, magnitude, or occurrence of PMI-associated effects on RNA integrity, and comparisons between studies have been complicated by evaluation of dissimilar PMI ranges, different tissue types, or a small number of donors. Results of the GTEx project indicate that the timing and magnitude of PMI effects on RIN are tissue-specific, findings that buttress those of published studies that have evaluated several different tissues for a PMI range of 24 hours or longer.7,18,19,36  While an overall effect of PMI on RIN was observed among GTEx specimens from the pilot study1  and among all pooled tissue specimens, further examination of individual tissue types revealed an early and progressive effect of PMI in pancreas and a weaker albeit significant effect in skeletal muscle, findings that support those previously reported by others.7,36  Other studies using data from GTEx tissue specimens have reported that PMI-associated mRNA degradation is influenced by tissue type and the gene targeted, as well as donor genotype.39  Further, analysis of RNA sequencing data from GTEx specimens revealed early alterations in gene expression beginning after a PMI of 6 hours in several tissue types; transcriptional responses included early, sustained, and peaked changes in expression that were tissue-specific.20  Importantly, the authors estimated that only 0.2% of genes per tissue were significantly correlated to PMI and that tissue-specific transcriptional profiles remained stable.20 

The timing and magnitude of PMI effects on the severity of autolysis were also tissue-specific in the GTEx sample set, which is supported by both well-accepted differences in autolysis rates due to different innate levels of intracellular enzymes among organs/tissue types and published studies. On average, autolysis scores were higher in kidney, pancreas, and other tissues with a high enzyme content, such as those involved in digestion. Similar findings that included early and robust histologic changes indicative of autolysis in postmortem pancreas and/or kidney specimens have been previously reported in both humans13  and mice.12,40  Comparatively, autolysis remained absent to mild in skeletal muscle and mild in tibial nerve specimens. Others have also reported that tissues with a high collagen content, such as skeletal muscle, display comparatively later and/or less severe changes.12,41  Indeed, the autolysis score was strongly correlated to PMI in pancreas specimens, whereas no such relationship was present in skeletal muscle. When a potential relationship between RIN and autolysis score was explored, a strong negative correlation was also observed in pancreas but not skeletal muscle. Studies investigating a potential relationship between RIN and autolysis are limited, although a significant association was identified in the cerebellar granule cell layer,21  and a concomitant increase in morphologic changes was observed with a decline in RIN with progressive PMI in mouse.40  Notably, autolysis rates can differ spatially within a tissue or organ,13  which could not be addressed here given the limited spatial sampling of each tissue collected for the GTEx project.

The reported association between rapidity of death (also referred to agonal phase) and RNA integrity has largely been investigated in brain specimens with conflicting results. Rapidity of death was recorded for each GTEx donor when possible using a modified Hardy scale as a way to annotate donors who received mechanical ventilation immediately prior to death (Hardy score = 0). When all tissue types were pooled, donors who received mechanical ventilation displayed a slightly lower mean autolysis score and a slightly higher mean RIN compared to donors who did not receive mechanical ventilation. However, results should be interpreted with caution given the skewed proportion of short (<8 hours) to long (≥8 hours) PMIs among donors when binned by Hardy score (Figure 4, B). The cause of this skewed distribution is unclear, but discussions prompted by the need for mechanical ventilation and end-of-life care may have allowed donor identification, eligibility assessment, and initiation of consent by next of kin to begin earlier than with other causes of death. Evaluation of tissue-specific differences in RIN relative to rapidity of death revealed weak negative correlations for 6 of the 40 tissue types examined, and no clear differences in the proportion of pancreas, nerve, or skeletal muscle specimens with RNA of low, intermediate, or high quality. Similarly, Hardy score was weakly and positively correlated with autolysis score only in stomach specimens. Others have also reported that rapidity of death was not associated with RNA integrity in cardiac muscle, skeletal muscle,36  or brain.36,42  Sheedy et al21  also observed no effect of agonal phase on the severity of autolysis in brain specimens. Interestingly, ventilator usage and sampling location were identified as 2 factors associated with the largest variance in expression among GTEx lung samples.35  Novel histologic lesions in the gastrointestinal tract were also observed among a subset of GTEx donors who received mechanical ventilation,23  suggesting that end-of-life treatments such as mechanical ventilation may affect tissues in multiple ways.

The GTEx project aimed to minimize variability in specimen collection and handling practices to the greatest extent possible by employing a single set of SOPs and using central processing laboratories; however, pragmatically, differences in donor characteristics and delays to autopsy were unavoidable. Effects of 2 such preanalytic factors, PMI and rapidity of death, were systematically evaluated in GTEx specimens; however, other preanalytic factors (such as donor age, body mass index, nonexclusionary clinical diagnoses, and manner of death) could also potentially influence histologic and molecular results. Further, retention of tissue samples with minor alterations that were observed during histologic review may be viewed as another limitation of the study given the likelihood that such changes are accompanied by deviations in gene expression. The occurrence of relatively minor but commonly encountered lesions associated with aging and chronic illness may blur the distinction between normal and abnormal in any large-scale study. In GTEx, such samples were analyzed, and all deviations were recorded in the electronic pathology report and included in dbGaP genomic sequence data submissions, providing the opportunity for others to study age-43–45  and chronic illness–46  related differences in gene transcription. Further, potential interactions between preanalytic factors not assessed in GTEx specimens cannot be excluded. Such challenges would affect any large and donor diverse study, and one might argue that the patterns of effect that were observed were large enough to transcend interdonor variability. Additionally, the comprehensive histologic review of each specimen allowed pathologists to extrapolate the scope of age-related morphologic changes, which are summarized in the companion paper.23  Other limitations in the analysis of the GTEx specimen cohort include that RNA integrity analysis was restricted to a single assay, RIN, that was selected based on evidence available during project initiation. While other RNA quality assays, such as DV200, have increased in popularity, RIN remains an informative metric of RNA integrity that mirrors results obtained with integrity assessments of amplificability in PAXgene-fixed tissue.27,47  The use of the PAXgene tissue fixation system may also be viewed as a study limitation given it is a less commonly used proprietary fixative. However, during the GTEx project, PAXgene fixative yielded RNA of superior quality to formalin, produced detailed histologic staining, and afforded the convenience of room-temperature fixation and storage (conclusions that have been reported by others48 ). A nonuniform distribution of donors across age, PMI, and Hardy score categories was also present among GTEx donors, which could both confound analysis of preanalytic factor-to-factor interactions and mask potentially significant effects of a single preanalytic factor. For example, while a significant correlation was not observed between Hardy score and RIN or autolysis score for many tissue types, the disproportionate distribution of donors across the Hardy score spectrum (Figure 3, B) could have obscured a potential relationship. Heterogenous effects of PMI or rapidity of death within a tissue specimen also cannot be excluded given the limited number of samples collected from an individual organ or tissue.

To the best of our knowledge, the GTEx project represents the largest standardized collection of normal tissues from postmortem donors to date, resulting in approximately 30 000 tissue aliquots from nearly 1000 donors. Findings presented here include insight on the challenges and benefits encountered during histologic review of a large cohort of diverse tissue types collected during a multicenter study, the importance of histologic evaluation as part of quality assessment measures, and tissue-specific differences in autolysis and RNA integrity and their susceptibilities to postmortem-associated preanalytic factors. Aspects relating to the ethical, legal, and social implications of the GTEx project have been explored in several studies, the results of which may prove informative to other studies relying on postmortem specimen donation.49–52  The specimens and data amassed during the GTEx project are a rich resource for the study of tissue-specific differences in genomes and gene expression that are unaltered by pathology, which has resulted in more than 1695 published articles by the scientific community to date. Information pertaining to the GTEx project, donors, tissues collected, histologic digital images, and all open access data analysis files (including expression count matrices and quantitative trait loci results), along with sample requests forms, are available through the GTEx portal (gtexportal.org). Histologic image files in Aperio format are available for download by contacting the National Cancer Institute (email: [email protected]). More than 50 requests for bulk downloads have been made by investigators worldwide, with a wide range of proposed analyses that include integrating histopathologic features with genomic signatures, training deep machine-learning models for image data analysis, and developing algorithms to recognize cancer-specific pathology features. Access to the raw DNA and RNA sequence-level files that include genotype data and metadata files (which include demographic, available clinical data, and pathology report annotations) can be requested through dbGaP (https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000424); to date, more than 2200 approved GTEx data requests have been made.

This resource would not have been possible without the generous donation of biospecimens and data by individuals and their families. We acknowledge the contributions of the many partners in the Genotype-Tissue Expression (GTEx) project, including the GTEx Consortium, Roswell Park Cancer Institute, National Disease Research Interchange and their network of contributing organ procurement organizations, the Miami Brain Bank, the Van Andel Institute, the Broad Institute, Leidos Biomedical Research, Inc, the National Institutes of Health (NIH) Common Fund, the National Human Genome Research Institute (NHGRI), the National Institute of Mental Health (NIMH), and the National Cancer Institute (NCI). We thank Francois Aguet, PhD, at the Broad Institute for his valuable insight and assistance with retrieving data from the LDACC, Maria M. Tomaszewski, MD, for her contributions as a project pathologist, and James Robb, MD, for his valuable contributions during protocol development. We gratefully acknowledge the leadership of Carolyn Compton, MD, PhD, in the early stages of this work.

1.
Carithers
LJ,
Ardlie
K,
Barcus
M,
et al.
A novel approach to high-quality postmortem tissue procurement: the GTEx project
.
Biopreserv Biobank
.
2015
;
13
(
5
):
311
319
.
2.
The GTEx Consortium. Human genomics.
The Genotype-Tissue Expression (GTEx) pilot analysis: multitissue gene regulation in humans
.
Science
.
2015
;
348
(
6235
):
648
660
.
3.
The GTEx Consortium.
The Genotype-Tissue Expression (GTEx) project
.
Nat Genet
.
2013
;
45
(
6
):
580
585
.
4.
Lockhart
NC,
Weil
CJ,
Carithers
LJ,
et al.
Development of a consensus approach for return of pathology incidental findings in the Genotype-Tissue Expression (GTEx) project
.
J Med Ethics
.
2018
;
44
(
9
):
643
645
.
5.
Biorepositories and Biospecimen Research Branch.
Genotype-Tissue Expression (GTEx) project SOPs
. https://biospecimens.cancer.gov/resources/sops/library.asp. Accessed September 15, 2023.
6.
Biorepositories and Biospecimen Research Branch.
National Cancer Institute biospecimen research database
.
Genotype-Tissue Expression (GTEx) project SOPs
. https://brd.nci.nih.gov/brd/sop-compendium/show/1742. Accessed December 1, 2023.
7.
Fan
J,
Khanin
R,
Sakamoto
H,
et al.
Quantification of nucleic acid quality in postmortem tissues from a cancer research autopsy program
.
Oncotarget
.
2016
;
7
(
41
):
66906
66921
.
8.
Birdsill
AC,
Walker
DG,
Lue
L,
Sue
LI,
Beach
TG.
Postmortem interval effect on RNA and gene expression in human brain tissue
.
Cell Tissue Bank
.
2011
;
12
(
4
):
311
318
.
9.
Liu
Y,
Gao
H,
Hu
Y,
Ding
J,
Ge
M,
Ye
Q.
Quality control system in an obstetrics and gynecology disease biobank
.
Biopreserv Biobank
.
2019
;
17
(
1
):
27
38
.
10.
Montanari
E,
Giorgetti
R,
Busardo
FP,
Giorgetti
A,
Tagliabracci
A,
Alessandrini
F.
Suitability of miRNA assessment in postmortem interval estimation
.
Eur Rev Med Pharmacol Sci
.
2021
;
25
(
4
):
1774
1787
.
11.
Preece
P,
Cairns
NJ.
Quantifying mRNA in postmortem human brain: influence of gender, age at death, postmortem interval, brain pH, agonal state and inter-lobe mRNA variance
.
Brain Res Mol Brain Res
.
2003
;
118
(
1–2
):
60
71
.
12.
Tomita
H,
Vawter
MP,
Walsh
DM,
et al.
Effect of agonal and postmortem factors on gene expression profile: quality control in microarray analyses of postmortem human brain
.
Biol Psychiatry
.
2004
;
55
(
4
):
346
352
.
13.
Cocariu
EA,
Mageriu
V,
Staniceanu
F,
Bastian
A,
Socoliuc
C,
Zurac
S.
Correlations between the autolytic changes and postmortem interval in refrigerated cadavers
.
Rom J Intern Med
.
2016
;
54
(
2
):
105
112
.
14.
Goggins
M,
Scott
JM,
Weir
DG.
Regional differences in protein carboxymethylation in post-mortem human brain
.
Clin Sci (Lond)
.
1998
;
94
(
6
):
677
685
.
15.
Gonzalez-Herrera
L,
Valenzuela
A,
Marchal
JA,
Lorente
JA,
Villanueva
E.
Studies on RNA integrity and gene expression in human myocardial tissue, pericardial fluid and blood, and its postmortem stability
.
Forensic Sci Int
.
2013
;
232
(
1–3
):
218
228
.
16.
Sherwood
KR,
Head
MW,
Walker
R,
Smith
C,
Ironside
JW,
Fazakerley
JK.
RNA integrity in post mortem human variant Creutzfeldt-Jakob disease (vCJD) and control brain tissue
.
Neuropathol Appl Neurobiol
.
2011
;
37
(
6
):
633
642
.
17.
De Paepe
ME,
Mao
Q,
Huang
C,
Zhu
D,
Jackson
CL,
Hansen
K.
Postmortem RNA and protein stability in perinatal human lungs
.
Diagn Mol Pathol
.
2002
;
11
(
3
):
170
176
.
18.
Walker
DG,
Whetzel
AM,
Serrano
G,
Sue
LI,
Lue
LF,
Beach
TG.
Characterization of RNA isolated from eighteen different human tissues: results from a rapid human autopsy program
.
Cell Tissue Bank
.
2016
;
17
(
3
):
361
375
.
19.
White
K,
Yang
P,
Li
L,
Farshori
A,
Medina
AE,
Zielke
HR.
Effect of postmortem interval and years in storage on RNA quality of tissue at a repository of the NIH NeuroBioBank
.
Biopreserv Biobank
.
2018
;
16
(
2
):
148
157
.
20.
Ferreira
PG,
Munoz-Aguirre
M,
Reverter
F,
et al.
The effects of death and post-mortem cold ischemia on human tissue transcriptomes
.
Nat Commun
.
2018
;
9
(
1
):
490
.
21.
Sheedy
D,
Harding
A,
Say
M,
Stevens
J,
Kril
JJ.
Histological assessment of cerebellar granule cell layer in postmortem brain; a useful marker of tissue integrity
?
Cell Tissue Bank
.
2012
;
13
(
4
):
521
527
.
22.
Hardy
JA,
Wester
P,
Winblad
B,
Gezelius
C,
Bring
G,
Eriksson
A.
The patients dying after long terminal phase have acidotic brains; implications for biochemical measurements on autopsy tissue
.
J Neural Transm
.
1985
;
61
(
3–4
):
253
264
.
23.
Branton
PA,
Sobin
L,
Barcus
M,
et al.
Notable histologic findings in a “normal” cohort: the National Institutes of Health Genotype-Tissue Expression project
.
Arch Pathol Lab Med
.
2024
. In press.
24.
Biorepositories and Biospecimen Research Branch.
BBRB-PM-0019 GTEx next of kin informed consent template. GTEx Standard Operating Procedures Library
. https://biospecimens.cancer.gov/resources/sops/library.asp. Accessed December 1, 2023.
25.
Biorepositories and Biospecimen Research Branch.
BBRB-PM-0003-F4 GTEx donor eligibility criteria form. GTEx Standard Operating Procedures Library
. https://biospecimens.cancer.gov/resources/sops/library.asp. Accessed December 1, 2023.
26.
Biorepositories and Biospecimen Research Branch.
BBRB-PR-0004 GTEx tissue processing procedure. GTEx Standard Operating Procedures Library
. https://biospecimens.cancer.gov/resources/sops/library.asp. Accessed December 1, 2023.
27.
Groelz
D,
Sobin
L,
Branton
P,
Compton
C,
Wyrich
R,
Rainen
L.
Non-formalin fixative versus formalin-fixed tissue: a comparison of histology and RNA quality
.
Exp Mol Pathol
.
2013
;
94
(
1
):
188
194
.
28.
Biorepositories and Biospecimen Research Branch.
BBRB-PR-0004-W1-G4 brain autopsy normal tissue collection
.
GTEx Standard Operating Procedures Library
. https://biospecimens.cancer.gov/resources/sops/library.asp. Accessed December 1, 2023.
29.
R Core Team
.
R: a language and environment for statistical computing
.
4.0.2 ed
.
Vienna, Austria
:
R Foundation for Statistical Computing
;
2020
.
30.
Gallego Romero
I,
Pai
AA,
Tung
J,
Gilad
Y.
RNA-seq: impact of RNA degradation on transcript quantification
.
BMC Biol
.
2014
;
12
:
42
.
31.
Houghton
JP,
Ervine
AJ,
Kenny
SL,
et al.
Concordance between digital pathology and light microscopy in general surgical pathology: a pilot study of 100 cases
.
J Clin Pathol
.
2014
;
67
(
12
):
1052
1055
.
32.
Snead
DR,
Tsang
YW,
Meskiri
A,
et al.
Validation of digital pathology imaging for primary histopathological diagnosis
.
Histopathology
.
2016
;
68
(
7
):
1063
1072
.
33.
Thrall
MJ,
Wimmer
JL,
Schwartz
MR.
Validation of multiple whole slide imaging scanners based on the guideline from the College of American Pathologists Pathology and Laboratory Quality Center
.
Arch Pathol Lab Med
.
2015
;
139
(
5
):
656
664
.
34.
Ash
JT,
Darnell
G,
Munro
D,
Engelhardt
BE.
Joint analysis of expression levels and histological images identifies genes associated with tissue morphology
.
Nat Commun
.
2021
;
12
(
1
):
1609
.
35.
McCall
MN,
Illei
PB,
Halushka
MK.
Complex sources of variation in tissue expression data: analysis of the GTEx lung transcriptome
.
Am J Hum Genet
.
2016
;
99
(
3
):
624
635
.
36.
Koppelkamm
A,
Vennemann
B,
Lutz-Bonengel
S,
Fracasso
T,
Vennemann
M.
RNA integrity in post-mortem samples: influencing parameters and implications on RT-qPCR assays
.
Int J Legal Med
.
2011
;
125
(
4
):
573
580
.
37.
Kap
M,
Oomen
M,
Arshad
S,
de Jong
B,
Riegman
P.
Fit for purpose frozen tissue collections by RNA integrity number-based quality control assurance at the Erasmus MC tissue bank
.
Biopreserv Biobank
.
2014
;
12
(
2
):
81
90
.
38.
van der Linden
A,
Blokker
BM,
Kap
M,
et al.
Post-mortem tissue biopsies obtained at minimally invasive autopsy: an RNA-quality analysis
.
PLoS One
.
2014
;
9
(
12
):
e115675
.
39.
Zhu
Y,
Wang
L,
Yin
Y,
Yang
E.
Systematic analysis of gene expression patterns associated with postmortem interval in human tissues
.
Sci Rep
.
2017
;
7
(
1
):
5435
.
40.
Hostetter
G,
Collins
E,
Varlan
P,
et al.
Veterinary and human biobanking practices: enhancing molecular sample integrity
.
Vet Pathol
.
2014
;
51
(
1
):
270
280
.
41.
Hostiuc
S,
Rusu
MC,
Manoiu
VS,
Vrapciu
AD,
Negoi
I,
Popescu
MV.
Usefulness of ultrastructure studies for the estimation of the postmortem interval. A systematic review
.
Rom J Morphol Embryol
.
2017
;
58
(
2
):
377
384
.
42.
Ervin
JF,
Heinzen
EL,
Cronin
KD,
et al.
Postmortem delay has minimal effect on brain RNA integrity
.
J Neuropathol Exp Neurol
.
2007
;
66
(
12
):
1093
1099
.
43.
Jia
K,
Cui
C,
Gao
Y,
Zhou
Y,
Cui
Q.
An analysis of aging-related genes derived from the Genotype-Tissue Expression project (GTEx)
.
Cell Death Discov
.
2018
;
4
:
26
.
44.
Yamamoto
R,
Chung
R,
Vazquez
JM,
et al.
Tissue-specific impacts of aging and genetics on gene expression patterns in humans
.
Nat Commun
.
2022
;
13
(
1
):
5803
.
45.
Yang
J,
Huang
T,
Petralia
F,
et al.
Synchronized age-related gene expression changes across multiple tissues in human and the link to complex diseases
.
Sci Rep
.
2015
;
5
:
15145
.
46.
Zeng
L,
Yang
J,
Peng
S,
et al.
Transcriptome analysis reveals the difference between “healthy” and “common” aging and their connection with age-related diseases
.
Aging Cell
.
2020
;
19
(
3
):
e13121
.
47.
Sanchez
I,
Betsou
F,
Culot
B,
et al.
RNA and microRNA stability in PAXgene-fixed paraffin-embedded tissue blocks after seven years’ storage
.
Am J Clin Pathol
.
2018
;
149
(
6
):
536
547
.
48.
Mathieson
W,
Marcon
N,
Antunes
L,
et al.
A critical evaluation of the PAXgene tissue fixation system: morphology, immunohistochemistry, molecular biology, and proteomics
.
Am J Clin Pathol
.
2016
;
146
(
1
):
25
40
.
49.
Siminoff
LA,
Traino
HM,
Mosavel
M,
Barker
L,
Gudger
G,
Undale
A.
Family decision maker perspectives on the return of genetic results in biobanking research
.
Genet Med
.
2016
;
18
(
1
):
82
88
.
50.
Siminoff
LA,
Wilson-Genderson
M,
Gardiner
HM,
Mosavel
M,
Barker
KL.
Consent to a postmortem tissue procurement study: distinguishing family decision makers’ knowledge of the Genotype-Tissue Expression project
.
Biopreserv Biobank
.
2018
;
16
(
3
):
200
206
.
51.
Siminoff
LA,
Wilson-Genderson
M,
Mosavel
M,
Barker
L,
Trgina
J,
Traino
HM.
Confidentiality in biobanking research: a comparison of donor and nondonor families’ understanding of risks
.
Genet Test Mol Biomarkers
.
2017
;
21
(
3
):
171
177
.
52.
Lockhart
NC,
Smith
AM,
Carithers
LJ,
Weil
CJ.
Genomic research with organs and tissues originating from transplant donors: ethical considerations for the GTEx project
.
IRB
.
2016
;
38
(
2
):
1
7
.

Author notes

Supplemental digital content is available for this article at https://meridian.allenpress.com/aplm in the March 2025 table of contents.

The Genotype-Tissue Expression (GTEx) project was supported by the Common Fund of the Office of the Director of the National Institutes of Health (NIH) and by the National Cancer Institute (NCI); National Human Genome Research Institute; National Heart, Lung, and Blood Institute (NHLBI); National Institute of Drug Abuse; National Institute of Mental Health; and National Institute of Neurological Disorders and Stroke. The work reported here was funded in whole or in part with federal funds from NCI NIH under contract HHSN261200800001E and from NHLBI and the NIH Common Fund under contract HHSN268201000029C.

The content here does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the United States Government.

Competing Interests

The authors have no relevant financial interest in the products or companies described in this article.

Supplementary data