Main

Xenotransplantation may offer a transformative solution to the worldwide organ shortage crisis1,2,3. To proceed to clinical studies, a clinically ready porcine donor must be engineered and its xenograft successfully tested in a nonhuman primate (NHP) model to assess its safety and efficacy.

Over the years, various genetically engineered porcine donors have been created and their kidneys transplanted into Old World monkeys (OWMs)4,5,6. Although these donors contributed to our understanding of molecular incompatibilities in xenotransplantation, they are not clinically ready. First, the donors were often created on a commercial pig breed whose heart and kidney sizes are too large for human application. Although elimination of growth hormone receptor gene expression could reduce organ sizes2,3, it comes with other undesired biological consequences7. Second, the donors were designed for testing in OWMs. They lacked the α-Gal (galactose-α-1,3-galactose) or the α-Gal and Sd(a) (Sia-α2.3-[GalNAc-β1.4]Gal-β1.4-GlcNAc) glycans but expressed the Neu5Gc (N-glycolylneuraminic acid) glycan to match with Neu5Gc expression in OWMs. However, in vitro analysis suggests that a human-compatible porcine donor should ideally have all three glycans eliminated to match with the absence of the three glycans in humans8,9. Although renal grafts derived from the porcine donors lacking these three glycans and carrying various human transgenes have been tested in OWMs, graft survival was short8 or not all human transgenes were expressed10. Third, the donors carried porcine endogenous retrovirus (PERV) sequences in their genome, which present a zoonotic risk, as PERV transmission to human cells in culture and their integration into the human genome have been demonstrated11,12.

Here we created a humanized porcine donor on the Yucatan miniature pig breed and transplanted porcine renal grafts lacking the three glycans with or without PERV knockout (retroviral inactivation (RI)) (referred to as 3KO.RI or 3KO), or 3KO with seven human transgenes with or without RI (referred to as 3KO.7TG.RI or 3KO.7TG, respectively) into cynomolgus monkeys (Macaca fascicularis). We show that a humanized porcine renal graft, combined with a clinically relevant immunosuppressive regimen, supported long-term NHP survival for up to 2 years (758 days).

Porcine molecular incompatibilities

Substantial molecular incompatibilities exist between pigs and humans. Of particular interest to xenotransplantation are cell-surface antigens and regulators of the complement cascade, coagulation pathway, and inflammation process13,14 (Supplementary Table 1).

Porcine cells display three major glycan antigens on their cell surface — α-Gal15, Neu5Gc16 and Sd(a)17 (Fig. 1a) — which are the products of the corresponding glycan synthesis genes, glycoprotein α-galactosyltransferase 1 (GGTA1), cytidine monophospho-N-acetylneuraminic acid hydroxylase (CMAH) and β-1,4-N-acetyl-galactosaminyltransferase 2 (B4GALNT2)/B4GALNT2-like (B4GALNT2L). In humans, GGTA1 (ref. 18) and CMAH19 evolved into pseudogenes and the α-Gal and the Neu5Gc epitopes are not expressed (Fig. 1a). After birth, humans develop antibodies to α-Gal and Neu5Gc, referred to as preformed antibodies, upon exposure to molecular mimics of these two antigens16,20,21. Although the human B4GALNT2 gene is functional22, its expression levels vary and naturally occurring mutations have been identified that correlate with the Sd(a-) phenotype23. A low level of Sd(a)-reactive antibodies has been detected in humans24. Similar to humans, OWMs carry preformed antibodies to α-Gal and Sd(a), but unlike humans, they lack preformed antibodies to Neu5Gc, as they possess a functional CMAH gene25 (Fig. 1a).

Fig. 1: Functional incompatibilities exist between porcine donor and primate recipient.
figure 1

a, IHC confirmed that the three glycan antigens are expressed in the WT porcine kidney, but have been eliminated from the 3KO.7TG (donor ID 21077) and the 3KO.7TG.RI (donor ID A9161) kidneys, a pattern similar to that of the human kidney. In comparison, OWMs express the Neu5Gc antigen. IHC analyses of 3KO were performed for all contralateral kidney samples included in this study and 3KO phenotypes confirmed for all. b, WT porcine KECs bound human and NHP preformed antibodies, whereas 3KO porcine KECs showed markedly reduced antibody binding. Each sample was run with three technical replicates. Error bars are s.d. c, The 3KO AECs bound significantly less IgG and IgM than the WT AECs, when incubated with 96 individual cynomolgus monkey serum samples. Note that 3KO AECs retained a substantial level of IgG and IgM binding. Statistical analysis was performed using Wilcoxon matched-pairs signed rank test. d, WT KECs showed significant C3b deposition in human serum, compared with human umbilical vein endothelial cells (HUVECs). Although markedly reduced, the 3KO KECs retained some level of C3b deposition (right panel). Error bars are s.e.m. e, WT KECs showed C3b deposition and 3KO KECs showed less deposition when incubated in the cynomolgus monkey serum. Error bars are s.e.m. f, Porcine KECs (WT or 3KO) did not produce aPC, whereas HUVECs readily produced aPC. For df, WT KEC (n = 1), 3KO KEC (n = 3) and HUVEC (n = 1). OD405, optical density at 405 nm. g, When incubated with human whole blood, WT (n = 1) and 3KO (n = 1) porcine KECs triggered coagulation, measured as TAT formation. Error bars are s.e.m. For dg, each point is a biological replicate examined over at least two independent experiments. With the exception of c, statistical analyses were performed using two-tailed unpaired Student’s t-tests. ****P < 0.0001, ***P < 0.001, **P < 0.01, *P < 0.05.

Source Data

When porcine cells are exposed to primate serum, the glycan antigens are recognized by primate preformed antibodies, leading to antibody-mediated rejection (AMR)13. In an antibody-binding assay, the porcine wild-type (WT) kidney endothelial cells (KECs) bound a high level of human IgG and IgM and binding was significantly reduced when the three xenoantigens were eliminated (3KO KECs) (Fig. 1b). Binding by OWM serum IgG and IgM was similar, although substantial residual antibody binding (especially IgM binding) was detected with the 3KO KECs (Fig. 1b). This is consistent with other reports that 3KO porcine cells possess additional xenoantigens recognized by OWM serum8,9. WT and 3KO aortic-derived endothelial cells (AECs) behaved similarly to the KECs when incubated with 96 individual cynomolgus monkey serum samples (Fig. 1c).

Antibody binding triggers complement activation, producing surface-bound C3b and soluble C3a. When incubated with human serum, human umbilical vein endothelial cells do not show C3b deposition (Fig. 1d), suggesting no antibody binding and/or complete mitigation of complement activation. When porcine WT KECs were incubated with human serum, significant C3b deposition was observed. Althoughthe 3KO KECs showed significant reduction in C3b deposition compared with the WT KECs, they retained substantial residual C3b deposition (Fig. 1d), suggesting that porcine complement regulators are less effective in mitigating human complement activation. When WT and 3KO porcine KECs were incubated with cynomolgus monkey serum, similar results were obtained, although higher residual C3b levels on 3KO porcine cells were observed, compared with those in human serum (Fig. 1e).

Under physiological conditions, thrombomodulin and endothelial protein C receptor (EPCR) are expressed on the endothelial cell surface and inhibit coagulation by enabling activated protein C (aPC)-mediated regulation26,27. When human thrombin and protein C were provided to human umbilical vein endothelial cells, aPC was readily generated (Fig. 1f). By contrast, aPC production was not observed when these reagents were supplied to porcine WT or 3KO KECs (Fig. 1f), suggesting that human thrombin and protein C are not compatible with porcine thrombomodulin and/or EPCR. Enhanced clotting of human whole blood ex vivo was observed, measured as thrombin–antithrombin (TAT) complex formation, probably due to the inability of the porcine cells to generate aPC (Fig. 1g).

A humanized porcine donor

The Yucatan miniature pig breed was chosen because its organ sizes are comparable to human organs28. In addition, pigs with OO blood type were selected to eliminate ABO blood-type incompatibilities29.

The pigs were engineered to carry 69 genomic edits, using the clustered regularly interspaced short palindromic repeats (CRISPR) and CRISPR-associated protein 9 (Cas9)-mediated nonhomologous end joining and homology-directed repair30,31, and recombinase-mediated cassette exchange32 (Fig. 2a, Extended Data Fig. 1a and Supplementary Table 2). These edits disrupted the three glycan synthesis genes (eight alleles; Extended Data Fig. 1b) (3KO), had a transgenic construct (referred to as Payload 15S (PL15S)) inserted hemizygously into the AAVS1 site (Extended Data Fig. 1a) (7TG), and inactivated the PERV elements (59 copies) (RI) (Extended Data Fig. 1c) carried in the Yucatan female cells Yuc25F.

Fig. 2: Yucatan porcine donor is engineered to carry 69 genomic edits.
figure 2

a, The porcine donor kidney, 3KO.7TG.RI, was engineered to eliminate three glycan antigens (3KO), overexpress seven human transgenes (PL15S) and inactivate PERV elements (RI) through three rounds of editing and cloning. The donor kidney, 3KO.7TG, carries 3KO and PL15S, without RI. KI, knock in; RMCE, recombinase-mediated cassette exchange. b, Reads from Nanopore long-read whole-genome sequencing of the 3KO.7TG.RI donor, A9161, were aligned to a custom chromosome carrying PL15S inserted at the AAVS1 genomic safe harbour site (top). Reads from Nanopore direct RNA-seq of A9161 kidney mRNA were aligned to the custom chromosome (bottom). All three transcription units were transcribed. c, All seven human transgenic proteins were detected in the contralateral (contra) kidney of A9161 recovered at transplantation (row 1), and the xenograft (xeno) kidney harvested at necropsy on post-transplantation day 176 (row 2). HRP, horseradish peroxidase; TM, thrombomodulin; Ms–HRP, goat anti-mouse secondary antibody conjugated with HRP; Rb–HRP, goat anti-rabbit secondary antibody conjugated with HRP. d, Three endothelial cell types were identified from kidney dissociated cell populations by single-cell RNA-seq, including endothelial cells (ECs) (PECAM1+PLVAPEHD3GATA5), glomerular endothelial cells (GECs) (PECAM1+EHD3+GATA5+) or fenestrated endothelial cells (FECs) (PECAM1+PLVAP+) (bottom), and transgene expression was examined among the three endothelial cell types (top). Mean log2-normalized unique molecular identifier counts were plotted against the three PL15S transcription units (ssUBC, ssEEF1A1 and CAG). UMAP, uniform manifold approximation and projection. e, The 3KO.7TG porcine donors (n = 3) showed normal measured glomerular filtration rate (mGFR) compared with age-matched WT Yucatan pigs (n = 4). Unpaired two-tailed Student’s t-test; error bars are s.e.m. Points are biological replicates and data are from one experiment.

Source Data

PL15S carries seven human genes (Supplementary Table 3), including CD46 and CD55 from the complement cascade, THBD and PROCR from the coagulation pathway, CD47, which is involved in innate immunity, and TNFAIP3 and HMOX1, which dampen ischaemia–reperfusion injury, apoptosis and inflammation. The transgenic construct was configured into a polycistronic design, in which two complementary DNA sequences were linked with a viral 2A sequence33 and the seven complementary DNAs split among three transcription cassettes (Extended Data Fig. 1a).

Next-generation sequencing was performed on the edited cells and/or the cloned pigs produced after each round of editing and cloning. Here we present data produced from the porcine donor, A9161, carrying the 3KO.7TG.RI genotype, whose kidney was transplanted into NHP recipient M6521 and achieved a graft survival time of 176 days. Long-read whole-genome sequencing showed that one copy of the intact PL15S sequence was inserted into intron 1 of the porcine PPP1R12C gene (orthologous to the human AAVS1 gene) (Fig. 2b, top). Direct RNA sequencing (dRNA-seq) of an A9161 kidney sample indicated that the three transcription units carried in PL15S were transcribed, with the expected transcription initiation, intron excision and mRNA polyadenylation (Fig. 2b, bottom). Among the three units, expression of the CAG and ssUBC units were higher, whereas expression of the ssEEF1A1 unit was at a lower level. The 3KO and RI genotypes were verified by sequencing of the PCR products encompassing the CRISPR–Cas9 target sites (Extended Data Fig. 1b,c and Supplementary Table 4).

RNA-seq and immunohistochemistry (IHC) were performed for all completed 3KO.7TG ± RI renal transplants, except for donor 21405 whose contralateral kidney biopsy sample was not available (n = 11). For RNA-seq, contralateral, biopsy and necropsy samples were analysed and showed that all human transgenes were expressed and, again, those under the CAG and ssUBC promoters were at a higher level, whereas those under the ssEEF1A1 promoter were at a lower level (Extended Data Fig. 2a). For IHC analysis (Supplementary Table 5), two samples from each renal transplant experiment, the contralateral kidney collected at transplant and the transplanted kidney procured upon necropsy, were analysed. As an example, IHC images from A9161 are shown (Fig. 2c). All seven transgenic proteins were detected in the contralateral kidney, in both the glomeruli and the tubular cells, and expression was maintained in the renal graft at necropsy on post-transplantation day 176. The IHC photomicrographs from all completed NHP studies were scanned and quantified (Extended Data Fig. 2b–e), which showed that all transgenic proteins were detected. Therefore, we conclude that transgene expression was durable.

Contralateral kidney tissues from two porcine donors, A9161 and 21077 (a 3KO.7TG donor), were dissociated into single cells, and single-cell RNA-seq was performed. Three KEC types were identified, including endothelial cells (PECAM1+PLVAPEHD3GATA5), fenestrated endothelial cells (PECAM1+PLVAP+) and glomerular endothelial cells (PECAM1+EHD3+GATA5+) (Fig. 2d). Transcripts from the CAG and ssUBC cassettes were readily detected in the three endothelial cell types, whereas ssEEF1A1 cassette transcripts were found in endothelial cells and fenestrated endothelial cells, but not in glomerular endothelial cells. This probably reflects the lower expression of the ssEEF1A1 cassette shown by direct RNA-seq (Fig. 2b). Nonetheless, both the human TNFAIP3 and HMOX1 proteins were detected in kidneys by IHC (Fig. 2c), and TNFAIP3 was detected by western blot (Extended Data Fig. 5c).

Given the relatively large number of genomic edits carried by the porcine donors, we wanted to see whether kidney function was compromised. The measured glomerular filtration rate of 3KO.7TG donors was not different from age-matched and gender-matched WT Yucatan pigs (Fig. 2e). Furthermore, when subjected to fluid challenge studies, both groups responded to the challenges similarly, providing further evidence that the 3KO.7TG kidneys functioned normally (Extended Data Fig. 1d).

Genomic edits confer protection

We isolated primary KECs from pigs (Extended Data Fig. 3a,b) and performed in vitro analysis. As expected, the 3KO KECs lacked α-Gal, Neu5Gc and Sd(a) (Extended Data Fig. 3c) and the 3KO.7TG ± RI KECs expressed human transgenes, as analysed by flow cytometry (Extended Data Fig. 4a, CD46, CD55, EPCR and thrombomodulin; Extended Data Fig. 5a, CD47) and by western blot (Extended Data Fig. 5c, TNFAIP3).

When incubated in human or cynomolgus monkey serum, the 3KO.7TG ± RI KECs exhibited significantly less C3b deposition than 3KO cells, suggesting that the transgene (or transgenes) impeded complement activation (Fig. 3a,b). When incubated with human or cynomolgus monkey serum for 45 min, WT cells were lysed, whereas the 3KO modifications alone almost completely abolished complement-dependent cytotoxicity of human serum, but not cynomolgus monkey serum (Fig. 3c and Extended Data Fig. 4b,c). This suggests that cynomolgus monkey serum has a stronger anti-porcine cytotoxic activity than human serum. When the incubation time was extended to 15 h, the 3KO modification was insufficient to fully protect the cells even from cell death by human serum, whereas 3KO.7TG ± RI KECs were protected against both human and cynomolgus monkey serum cytotoxicity, beyond the protection afforded by 3KO (Extended Data Fig. 4b). The contribution of the CD46 and CD55 transgenes was verified by blocking antibodies (Fig. 3d and Extended Data Fig. 4d,e). Furthermore, when CD46, CD55 or both were expressed in 3KO KECs, each was functionally competent to mitigate complement activation (Fig. 3e). Collectively, these data demonstrate that transgenic human CD46 and CD55 proteins regulate complement activity when expressed on porcine KECs. By including both transgenes, we mimic the naturally built-in redundancy, which may render the system more resilient in the case that one of the regulators is lost by shedding (CD46)34 or by enzymatic cleavage (CD55)35 during inflammation or tissue injury.

Fig. 3: Human transgenes confer protection.
figure 3

a,b, In human (a) or cynomolgus monkey (b) serum, deposition of C3b on 3KO.7TG ± RI KECs was further reduced. The dashed lines indicate average C3b deposition (serum + EDTA). c, In human or cynomolgus monkey serum, 3KO.7TG ± RI KECs were protected from complement-dependent cytotoxicity. 3KO KECs showed residual complement-dependent cytotoxicity in cynomolgus monkey serum. d, The 3KO.7TG ± RI KECs mitigated C3b deposition when human CD46, CD55 or both were not blocked by antibodies. NHS, normal human serum. e, 3KO KECs carrying CD46, CD55 or CD46CD55CD47 polycistronic genes regulated C3b deposition in human or cynomolgus monkey serum. The dashed lines indicate average C3b deposition from EDTA-treated serum. f, The 3KO.7TG ± RI KECs readily generated aPC. aPC production was reduced when EPCR or thrombomodulin (TM) was blocked with antibodies, with blocking of thrombomodulin being more effective. The dashed line indicates average aPC production without cells. g, 3KO.7TG ± RI KECs regulated TAT formation in whole blood. The dashed line indicates baseline TAT in donor blood. h, Human CD47 (hCD47) on 3KO.7TG ± RI KECs signalled through SIRPα on human Jurkat cells. WT KECs (n = 1), 3KO KECs (n = 1), 3KO.7TG ± RI (n = 2 for 3KO.7TG and n = 1 for 3KO.7TG.RI). mAb, monoclonal antibody; RLU, relative light unit. i, 3KO.7TG ± RI KECs were protected from human TNF-induced caspase 3/7 activation. j, 3KO kidney cortex-derived cells expressing TNFAIP3, HMOX1 or TNFAIP3HMOX1 polycistronic genes were protected from human TNF-induced caspase 3/7 activation. In ad,fi, WT (n = 1), 3KO (n = 3), 3KO.7TG (n = 3, solid blue circles), 3KO.7TG.RI (n = 2, open circles), human controls (HC) and human umbilical vein endothelial cells (n = 1, grey triangle). In ad,f,g, human glomerular microvascular endothelial cells (n = 1, grey inverted triangle). For ad,fi, data are from biological replicate, repeated at least twice. For e,j, data are from a technical replicate. Error bars indicate s.e.m. Statistical analyses were two-tailed, paired (3KO.7TG ± RI with versus without blocking antibodies (f)) and unpaired (all other comparisons) Student’s t-tests. ****P < 0.0001, ***P < 0.001, **P < 0.01; NS, not significant.

Source Data

Unlike WT or 3KO porcine cells, 3KO.7TG ± RI KECs readily produced aPC (Fig. 3f), reduced TAT formation (Fig. 3g) and regulated coagulation of human whole blood as efficiently as human control cells (Extended Data Fig. 4h). Using blocking antibodies (Extended Data Fig. 4f), we showed that both EPCR and thrombomodulin contributed to aPC production (Fig. 3f), with thrombomodulin having a more prominent effect. In addition, 3KO KECs overexpressing EPCR, thrombomodulin or both were engineered and showed that both EPCR and thrombomodulin contributed to aPC production, with thrombomodulin having a more significant effect in this experimental system (Extended Data Fig. 4g). Although both thrombomodulin and EPCR contribute to aPC production, their mechanisms of action differ36. Thrombomodulin binds to thrombin and functions as a cofactor in the thrombin-induced activation of protein C, whereas EPCR binds to protein C and presents it to the thrombomodulin–thrombin activation complex. Given the synergistic effect of the two proteins, we reasoned that coagulation may be better regulated when both are provided.

Porcine CD47 does not engage the human SIRPα receptor effectively37. Using a SIRPα reporter cell line, we found that 3KO.7TG ± RI KECs expressing transgenic human CD47 (Extended Data Fig. 5a) activated the SIRPα signalling pathway, and pre-incubation with the anti-human CD47 antibody blocked SIRPα signalling in a dose-dependent manner (Fig. 3h). In addition, the capacity of human CD47 to engage NHP SIRPα receptors was demonstrated in a binding assay using a human CD47 fusion protein to stain monocytes from cynomolgus monkey expressing endogenous SIRPα (Extended Data Fig. 5b).

Finally, activation of innate immune cells in the xenograft induces inflammation and apoptosis of endothelial cells13. To improve tissue resilience, human TNFAIP3 and HMOX1 were expressed in 3KO.7TG ± RI kidneys (Fig. 2c). Western blot analysis confirmed TNFAIP3 expression in 3KO.7TG ± RI KECs (Extended Data Fig. 5c). The effect of transgenic TNFAIP3 and HMOX1 expression on apoptosis was assessed in a caspase 3/7 assay, which showed that transgenic protein levels in 3KO.7TG cells were sufficient to reduce caspase 3/7 activation following human TNF treatment, compared with 3KO KECs (Fig. 3i). Furthermore, kidney cortex-derived cells overexpressing TNFAIP3, HMOX1 or both effectively blocked TNF-induced caspase 3/7 activation in vitro (Fig. 3j). Although overlapping in their anti-apoptotic activities, TNFAIP3 and HMOX1 achieve this outcome by their unique biological functions and dampen inflammation under distinct circumstances. TNFAIP3 is a deubiquitinating enzyme and inhibits TNF-mediated apoptosis38, whereas HMOX1 is involved in the initial step of haem degradation and has been shown to have antioxidant and anti-inflammatory effects38,39. We included both TNFAIP3 and HMOX1 in our donors, as both transgenes have been shown to provide benefit in vitro and in vivo40,41.

Engineered porcine kidney supports life

A cohort of cynomolgus monkeys was screened for porcine-reactive preformed antibody binding. In general, monkeys with a lower antibody binding to 3KO AECs or KECs were selected (Extended Data Fig. 6a–d).

Given that NHPs are expensive, highly regulated and limited in availability, it was not possible to conduct a statistically powered experiment. Therefore, the number of NHP transplants per porcine donor genotype group was empirically determined, based on historical data reported in literature and what could be reasonably achieved. The pairing of a porcine donor with an NHP recipient was dictated by availability. In addition, the study was not blind for all involved. The recipients were administered an immunosuppression regimen of induction therapy with B and T lymphocyte depletion, maintenance therapy with anti-CD154 antibody and mycophenolate mofetil, and a brief post-transplant course of tacrolimus and steroids (Supplementary Fig. 1). A genetically engineered porcine kidney was transplanted, concurrently with nephrectomy of the two native kidneys of the cynomolgus monkey.

Given the similar performance of KECs isolated from 3KO.7TG or 3KO.7TG.RI donors in functional assays (Fig. 3 and Extended Data Figs. 35), we analysed transplants performed with 3KO kidneys with and without RI as one group and 3KO.7TG and 3KO.7TG.RI kidneys as a second group. Survival of the six 3KO ± RI kidney transplant recipients was short, with end of study at days 4 and 6 (renal failure), 21 (disseminated intravascular coagulation), and 26, 35 and 50 (severe oedema and proteinuria) (Table 1). By contrast, the 3KO.7TG (n = 8) and 3KO.7TG.RI (n = 7) transplants achieved significantly longer graft survival than the 3KO ± RI xenografts (n = 6) (median survival time of 176 days versus 24 days; P = 0.026, log-rank test) (Table 1 and Fig. 4a). The filtration of metabolites, such as creatinine, by the single transplanted porcine kidney was sufficient to compensate for the lack of two native kidneys (Fig. 4b), as observed routinely in human renal allotransplantation. Other parameters, including serum albumin, serum potassium and blood platelet counts, generally remained within normal range, except when associated with renal failure (serum albumin and potassium) (Supplementary Fig. 5).

Table 1 Renal grafts support life in pig-to-NHP xenotransplantation
Fig. 4:  Engineered porcine kidneys support life in cynomolgus monkeys.
figure 4

a, Significantly improved survival probability (P = 0.026) for recipients transplanted with 3KO.7TG ± RI renal grafts (n = 15) compared with 3KO ± RI (n = 6). Statistical analysis was performed using a log-rank test. b, Serum creatinine levels generally remained within normal range, except when associated with graft failure. Creatinine levels above 6 mg dl−1 meet criteria for euthanasia. Dashed lines indicate normal range of serum creatinine levels. c, Periodic acid–Schiff staining of a kidney biopsy sample derived from recipient M2420, carrying a 3KO.7TG renal graft, at POD 502, appears normal. d,e, Blood vessels (d) and glomerulus (e) of M2420 renal graft biopsy at POD 502 appear normal. f, C4d-positive staining in glomeruli, but not in peritubular capillaries, was observed in M2420 renal graft biopsy at POD 502.

Source Data

Rejection was assessed by histopathological examination, based on the Banff classification of renal allograft pathology42, with modifications for xenotransplantation (Table 1 and Extended Data Table 1). Five out of the six 3KO ± RI kidneys showed evidence of acute tubular injury (ATI), a feature not included in the Banff criteria but representing general kidney injury. In addition, antibody-mediated rejection (AMR)/thrombotic microangiopathy (TMA) (M12121), TMA (CY1061, CY1062 and MB1027) and T cell-mediated rejection (TCMR) (M11521) were also observed. In 3KO.7TG ± RI kidney recipients, those who developed renal failure exhibited TMA with or without AMR at time of euthanasia and only two showed evidence of ATI. Unlike allotransplantation, T cell-mediated rejection was not a major pathology in xenotransplantation, with only one graft loss meeting Banff criteria for TCMR. It remains to be determined whether the current Banff scoring system developed for allotransplantation in humans can be satisfactorily applied to xenotransplantation, as rejection mechanisms may differ43. In allotransplantation, TMA usually reflects an antibody-mediated or drug-related process, whereas in xenotransplantation, substantial species coagulation or complement regulatory protein incompatibility may contribute to TMA, even in the absence of de novo donor-specific antibody (dnDSA). With these considerations, pathology samples were assessed with a ‘xenotransplantation-adapted’ criteria (Extended Data Table 1 and Supplementary Table 6). These xenotransplantation-adapted criteria represent our effort at informative and useful scoring and diagnosis in renal xenotransplantation.

Lymphocyte depletion in blood was demonstrated by measuring circulating B cell and T cell counts (Extended Data Fig. 7). Although B cell depletion was robust, dnDSA reactive to 3KO endothelial cells was detected in some animals over time (8 of 18 transplants; Table 1 and Extended Data Fig. 8) and pathological features of antibody-mediated graft injury were observed. For recipient M2420 with the longest xenograft survival time (758 days), a protocol biopsy showed mostly normal kidney histology at day 502 after transplant, with patchy fibrosis (Fig. 4c–e and Extended Data Fig. 9a). However, C4d staining was observed in the glomeruli, suggesting C4 activation (Fig. 4f and Extended Data Fig. 9b). C4 activation is upstream of CD46 and CD55 intervention and positive C4d staining may be expected in the 3KO.7TG ± RI kidney samples in the presence of dnDSA. Representative histopathological photomicrographs from necropsy samples with survival times of 9 (M10619), 176 (M6521) and 758 (M2420) days are provided in Extended Data Fig. 9c–e. Among the long-term survivors, simple appearing, benign cysts in the kidney were observed. Their aetiology is unknown and currently under investigation.

In NHPs with dnDSA, it is possible that the current immunosuppressive regimen was initially efficacious but eventually failed to prevent the development of porcine-specific humoral responses. TMA and AMR in the long-term survivors were often associated with infections or biopsy procedures, which might have triggered an immune response. Although these observations can inform the design of immunosuppressive regimens for clinical studies (for example, consideration of depletion with an anti-CD19 antibody and desensitization or salvage therapy with plasmapheresis or targeted plasma cell therapies), approaches that can be applied in the NHP model are limited. Of note, agents targeting the CD40–CD40L pathway have become standard of care in NHP xenotransplantation studies. Although these agents have not received FDA approval, several are in clinical development and will serve as a cornerstone in the immunosuppressive regimen, along with FDA-approved drugs currently used in clinical kidney transplantation.

Discussion

In this study, we describe a porcine donor carrying 69 genomic edits, with expression and function of all 7 human transgenes. Although a porcine donor carrying a CMAH knockout is thought not to survive in OWMs considering complication from a CMAH knockout8,9, renal grafts derived from the 3KO.7TG ± RI porcine donor supported life long-term in cynomolgus monkeys, up to 758 days.

The variable graft survival time may be inherent to kidney xenotransplantation and/or unique to the OWM model. Although 3KO substantially reduced the level of pre-formed anti-porcine antibody binding, minor xenoantigens remain and contributed to residual antibody binding (Fig. 1b). Furthermore, complement-dependent cytotoxicity activity was higher in cynomolgus monkey serum than in human serum (Fig. 3c and Extended Data Fig. 4b,c). In addition, data suggest that CMAH inactivation may produce a novel antigen xenogenic to the OWMs, which may be a target of preformed antibody binding, leading to complement activation in OWM serum8,9. It is worth noting that in a clinical setting, the complication around CMAH knockout unique to the OWMs will not be relevant, as it will be a match with the CMAH pseudogene genotype of the human recipient.

One advantage of xenotransplantation, as compared with allotransplantation, is the opportunity to genetically engineer a donor organ. Therefore, to promote graft survival, the burden may be shifted from a heavy immunosuppression regimen on the recipient to a more optimal graft donated from a genetically engineered porcine donor. Given the substantial molecular incompatibilities between the two species, the inclusion of additional human transgenes may be considered, such as TFPI, CD39 (also known as ENTPD1), HLA-E and PDL1 (also known as CD274)14,44. To recapitulate the pattern and level of expression of the porcine endogenous genes, in situ knock-in, in which the human coding sequence replaces the porcine orthologous gene, could be considered. This is particularly pertinent for THBD and PROCR, which are normally expressed in endothelial cells. We recognize that in addition to the glycan antigens, a vast array of porcine proteins may also be xenogenic17 and it may not be possible to identify and eliminate them all. Ultimately, a genetically engineered porcine model, with an immune tolerance feature45, may be the goal.

The successful proof-of-principle study achieved in this study brings us closer to clinical testing of porcine renal grafts for human transplantation.

Methods

Assembly of transgenic constructs

The landing pad construct carries the human EEF1A1 promoter driving expression of the mTagBFP2 marker gene, flanked by a pair of loxP/lox2272 sites. The left homology arm of 1,469 bp (chromosome 6 coordinates, from 59,347,343–59,345,875, susScr11) and the right homology arm of 1,260 bp (chromosome 6 coordinates, from 59,345,874 to 59,344,615, susScr11) were amplified from genomic DNA isolated from the Yucatan breed and placed 5′ or 3′ to the loxP/lox2272 sites. The PL15S transgenic construct was assembled by yeast homologous recombination46. In brief, the human coding DNA sequences (Supplementary Table 3), promoter, terminator and polyadenylation sequences were arranged into one of the three polycistronic transcription units, which were further arranged into a linear DNA molecule in a convergent or divergent configuration (Extended Data Fig. 1).

CRISPR–Cas9 guide RNA design

The R library DECIPHER47 was used to design guide RNAs targeting the sequence encoding the catalytic core of the pol enzyme from the PERV element (Supplementary Table 2). To inactivate the four genes involved in synthesizing the three major glycan antigens, α-Gal, Neu5Gc and Sd(a), we used one single guide RNA (sgRNA) per gene, targeting the GGTA1, CMAH or B4GALNT2/B4GALNT2L gene. To insert the landing pad DNA into the AAVS1 genomic locus, we used one guide RNA targeting the AAVS1 locus. Guide sequences are provided in Supplementary Table 2. All sgRNAs were synthesized and provided by Synthego.

CRISPR–Cas9-mediated nonhomologous end joining and homology-directed repair mutations

An sgRNA was incubated with the Cas9 enzyme (A36496, Thermo Fisher) to form the ribonucleoprotein (RNP) complex immediately before use, according to the manufacturer’s instructions. To elicit knockouts of the three xenoantigen genes and insertion of the landing pad in a multiplexed reaction, the AAVS1 landing pad donor plasmid was added to the RNP complexes before electroporation. To inactivate the PERV elements, the three sgRNAs were complexed with the Cas9 protein to form RNPs. Electroporation was performed with the Neon Transfection System 100 µl Kit (MPK10096, Thermo Fisher).

Generation of the porcine donors carrying 3KO, PL15S insertion into the AAVS1 site (7TG) and RI

A total of 1 × 106 ear-punch-derived cells (EPDCs) were electroporated (MPK10096, Thermo Fisher) with the four RNPs targeting the GGTA1, CMAH and B4GALNT2/B4GALNT2L genes, and the AAVS1 locus, together with the AAVS1 landing pad donor plasmid, and five days later, cells were sorted into single cells gated on the absence of the α-Gal glycan (isolectin B4, FITC conjugate, ALX-650-001F-MC05, Enzo Life Sciences) and the presence of the mTagBFP2 marker gene and placed into individual wells of a 96-well plate. Clonal populations of the cells were subsequently genotyped to identify those that carried the correct edits of 3KO and successful landing pad insertion into the AAVS1 site (Supplementary Table 4). Edited cells were then used as nuclear donors in a somatic cell nuclear transfer (SCNT) experiment to produce pigs carrying these edits.

EPDCs carrying 3KO and the landing pad inserted at the AAVS1 site were isolated and electroporated with the PL15S transgenic construct, along with the Cre recombinase mRNA (130-101-113, Miltenyi Biotec), to enable recombinase-mediated cassette exchange. Subsequently, cells were sorted into single cells gated on cell-surface expression of the genes carried on the payload (CD46, CD55, PROCR, THBD or CD47), and placed into single wells of a 96-well plate. Clonal populations of cells were genotyped to identify those that carry successful replacement of the landing pad with the PL15S sequence (Supplementary Table 4). These edited cells were used in an SCNT experiment and cloned into pigs.

To determine the number of copies of the PERV elements carried in the Yucatan genome, droplet digital PCR was performed as previously described48. Analysed with an assay designed against the pol gene, the Yucatan 25 female line (Yuc25F) was found to carry 59 copies of the sequence. To inactivate the PERV elements, EPDCs carrying 3KO and PL15S inserted at the AAVS1 site were derived and electroporated with the three RNPs targeting the catalytic core of the reverse transcriptase activity of the pol gene (Supplementary Table 2), and single cells were sorted into single wells of a 96-well plate. The clonal populations of cells were genotyped by amplicon sequencing (Supplementary Table 4) and those carrying indel mutations on all copies of the PERV elements were identified. The edited cells were used in an SCNT experiment and cloned into pigs.

Porcine donor production by somatic cell nuclear transfer

Gene-edited cells were cloned into pigs by SCNT49 and cloning was performed by eGenesis Wisconsin and Precigen Exemplar. Animal cloning was performed under Institutional Animal Care and Use Committee-approved protocols (eGenesis Wisconsin protocol HF2020-01, approved 24 November 2020; Precigen Exemplar protocol MRP2018-003, approved 21 June 2018). All resulting porcine Yucatan donors (Sus scrofa domesticus) were female Yucatans. All donor production strictly followed the Guide for the Care and Use of Laboratory Animals (National Research Council of National Academies), particularly the 3R principles.

IHC staining and analysis of transgene expression

Expression of the human proteins (EPCR, thrombomodulin, TNFAIP3, HMOX1, CD46, CD55 and CD47) was assessed from formalin-fixed, paraffin-embedded tissue sections of 8-week-old WT and transgenic Yucatan porcine kidney samples based on standard (tyramide signal amplification) protocols using Cy5+-tyramide as detection reagent. After all targets were detected, tissue sections were counterstained with Hoechst 33258 and mounted with ProLong Glass antifade mountant. Stained tissue sections were imaged using a Zeiss Axio Scan.Z1 automated whole-slide fluorescence scanner using the same scanning parameters for each tissue. Images were generated using the Zeiss Zen Blue 3.4 image analysis software. A list of reagents is provided in Supplementary Table 5.

Mean fluorescent intensities (MFIs) of the transgene proteins in the whole kidney tissues, glomeruli, tubules and blood vessels of positive and negative controls were measured using the Zeiss ZEN Blue 3.4 image analysis software. Positive controls consisted of samples stained with primary antibodies specific to the human transgene proteins, goat anti-mouse–horseradish peroxidase (HRP) or goat anti-rabbit–HRP conjugate and Cy5+-tyramide. Negative controls consisted of samples incubated with the mouse–HRP or rabbit–HRP conjugate and Cy5+-tyramide only.

Whole-tissue and tissue biopsy measurements were done by drawing around the contour of the entire tissue. The MFI values correspond to the pixel intensities inside the contour as calculated by the software. For the glomerular and blood vessel MFI measurements in whole tissue, 20 glomeruli and 8 blood vessels were selected randomly with equal distribution within the tissue, and contours were drawn to define the structures. For small tissue biopsies, 5–10 glomeruli and 3–5 blood vessels were selected. Average MFI values for the glomeruli and blood vessels (per whole tissue or tissue biopsy) were calculated. Tubular MFI measurement was done using 20 rectangles (measured by the software as 500 × 500 pixels) and 5–10 rectangles (measured by the software as 300 × 300 pixels) in whole tissue and tissue biopsy, respectively. The average MFI values for the 20 rectangles (per tissue) or 5–10 rectangles (per tissue biopsy) were calculated. The reported MFI values in the bar graphs correspond to the normalized average MFI (that is, specific signal) of the 11 samples. Normalized average MFI is defined as the average MFI values of the positive controls minus the average MFI values of the negative controls (that is, nonspecific signal).

IHC staining for 3KO in kidneys

Formalin-fixed, paraffin-embedded tissue sections of kidney samples from Yucatan porcine, human, cynomolgus, rhesus and baboon were processed as described above using the 1X Thermo Fisher citrate buffer for heat-induced epitope retrieval in a pretreatment (PT) module. After heat-induced epitope retrieval, tissue sections were blocked in TBS plus 5% goat serum for 30 min followed by a 2-h incubation with a mixture of binding reagents in TBS with 5% goat serum. The binding reagent mix consisted of 1:100 dilution of isolectin B4-FITC (detecting the α-Gal antigen), 1:100 dilution of chicken anti-Neu5GC antibody (detecting the Neu5GC antigen) and 1:250 dilution of DBA-biotin (detecting the Sd(a) antigen). Tissue sections were washed with TBS-T three times for 3 min each time and incubated with a mixture of a 1:1,000 dilution of goat anti-chicken Alexa Fluor 647 and a 1:1,000 dilution of streptavidin-Alexa Fluor 568 in TBS supplemented with 5% goat serum. Nuclear staining, tissue mounting with ProLong Glass antifade mountant, imaging and image analysis were performed as described above.

Measured glomerular filtration rate analysis

Four-month-old WT and the 3KO.7TG Yucatan swine received a single intravenous dose of Omnipaque 300 Iohexol at 64.7 mg kg−1 (00407141359, GE Healthcare). Blood samples were collected at 5, 15 and 30 min and 1, 2, 3, 6, 8 10 and 24 h post-dosing, and plasma was separated using K2EDTA. Iohexol concentrations were measured using high-performance liquid chromatography with ultraviolet spectroscopy (HPLC-UV). Various pharmacokinetic parameters including clearance (ml min kg−1) were calculated using a two-compartmental model (Phoenix WinNonlin, version 8.1 software). Body surface area (BSA) was calculated as BSA (m2) = 9 × BW2/3/100, and measured glomerular filtration rate (ml min m2) was calculated as Iohexol clearance normalized to BSA50. Data were plotted and statistics were calculated using GraphPad Prism v8.2.0.

Primate anti-porcine IgG and IgM analysis

Endothelial cells from 3KO porcine animals without human transgenes were used to measure anti-porcine IgG and IgM antibodies in heat-inactivated serum obtained from cynomolgus recipients before and after transplantation, along with pools of cynomolgus serum from 96 animals, and human serum from at least 100 healthy donors (SeraCare Life Sciences). Each endothelial cell sample (1 × 105 cells per test) was incubated with serum diluted 1:4 in 1× PBS with 1% BSA at 4 °C for 30 min. Cells were washed and incubated at 4 °C for 30 min with Alexa Fluor 488 conjugated F(ab′)2 anti-human IgG (109-546-098, Jackson ImmunoResearch) or Alexa Fluor 647 conjugated F(ab′)2 anti-human IgM (109-606-129, Jackson ImmunoResearch) secondary antibody diluted 1:100. The samples were fixed in 4.2% paraformaldehyde, acquired within 3 days on a FACSymphony A3 (BD Bioscience) and analysed using Flow Jo software v10.6.1 (Flow Jo LLC). MFI levels of IgG and IgM were evaluated in duplicate. MFI data were plotted and statistics were calculated using GraphPad Prism v8.2.0.

C3b deposition assay

Endothelial cells (50,000 cells per well) were seeded in a 96-well plate in serum dilution buffer (SDB) (1× annexin V buffer (51-66121E, BD Pharmingen), 1 mM MgCl2 (68475, Sigma) and 1% BSA (A9576, Sigma)). Pooled normal human serum (NHS, Complement Technology) or cynomolgus monkey serum (NHP01SRM, BioIVT) diluted in SDB were added to appropriate wells at a final concentration of 25% and incubated for 30 min at 37 °C. For negative controls, cells were treated with 25% serum containing 10 mM EDTA (15575038, Thermo Fisher) to inactivate complement. After incubation, cells were washed and stained with phycoerythrin-conjugated anti-C3b antibodies at a 1:100 dilution (846104, BioLegend) and Ghost Dye Red 780 viability dye at a 1:500 dilution (13-0865, Tonbo Biosciences) for 30 min at 4 °C in the dark. Cells were washed twice, immediately acquired on a BD FACSymphony A3 cytometer and analysed in Flow Jo. C3b deposition was plotted as MFI and statistics were calculated using GraphPad Prism v8.2.0.

Complement-dependent cytotoxicity assay

One day before the assay, endothelial cells (3,000 cells per well) were seeded in a 96-well plate (3595, Corning) in endothelial cell base medium. The next day, adherent cells were washed once with SDB, treated with 25% diluted sera (described above) containing 250 nM Cytotox Red viability dye (4632, Essen Biosciences), immediately placed in an Incucyte SX5 live-cell analysis system (model S3, Sartorius) and incubated at 37 °C in a CO2 incubator for the durations indicated in the figure legends. The number of Cytotox Red-positive cells was counted by the Incucyte software, and total cell counts were determined manually from phase contrast images. Complement-dependent cytotoxicity was calculated by normalizing the number of Cytotox Red-positive cells to the total number of cells. Data were plotted as percent Cytotox Red-positive cells, and statistics were calculated using GraphPad Prism v8.2.0.

aPC assay

The day before the assay, endothelial cells (20,000 cells per well) were seeded in a 48-well plate (FB012930, Thermo Fisher) in endothelial cell base medium. The next day, adherent cells were washed once with assay buffer (10 mM Tris HCl (15567-027, Thermo Fisher), 150 mM NaCl (S5886, Sigma), 5 mM CaCl2 (C1016, Sigma), 0.1% BSA (A9576, Sigma), pH 7.5), and then, where applicable, incubated with 40 μg ml−1 RCR-252 (2× final concentration; MA5-33375, Thermo Fisher) and/or 4 μg ml−1 PBS-01 (2× final concentration; ab6980, Abcam) for 1 h at room temperature in assay buffer. After incubation and without removing the blocking antibodies, cells were treated with 0.1 U ml−1 thrombin (605190, Sigma) and 150 nM protein C (539215, Sigma), both diluted in assay buffer, for 60 min at 37 °C. After incubation, 2 U ml−1 hirudin (H0393, Sigma) diluted in assay buffer was added to quench thrombin activity and the plate was incubated for 5 min at 37 °C. The solutions from each well were transferred to a 96-well plate, alongside a serial dilution of purified human aPC (HCAPC-0080, Prolytix) diluted in assay buffer to produce a standard curve. Spectrozyme PCa chromogenic substrate (336, Biomedica Diagnostics), diluted to 1 mM in imidazole buffer (0.1985 g ml−1 imidazole (I5513, Sigma), 0.03535 g ml−1 Tris (BP152, Thermo Fisher), 0.12675 g ml−1 NaCl and 250 mM HCl, pH 8.5), was added and absorbance read at 405 nm every 30 s for 15 min on a microplate reader (FilterMax F5, Molecular Devices). Initial velocity of the reaction was calculated (slope of the initial linear part of the curve), and aPC concentrations were determined using the aPC standard curve. Concentration data were plotted, and statistics were calculated using GraphPad Prism v8.2.0.

Whole-blood TAT complex assay

The day before the assay, endothelial cells (75,000 cells per well) were seeded in a 24-well plate (353226, Corning) in endothelial cell base medium. Fresh whole blood was collected in-house by a phlebotomist (Quadrant Health) in a Vacutainer serum collection tube (367820, BD Biosciences) containing 0.5 U ml−1 heparin (H3393, Sigma), 225 μl added to the adherent cells after a wash with 1× PBS and incubated at 37 °C on an Orbitron plate rocker (201100, Boekel Scientific) for 40 min. After incubation, non-clotted blood was transferred to Eppendorf tubes and centrifuged at 1,500g for 10 min. For baseline TAT concentration, whole blood was centrifuged at the same time blood was being added to cells. Plasma was collected and frozen on dry ice until use. Plasma samples were used in a TAT complex ELISA (ab108907, Abcam) according to the manufacturer’s instructions with TAT concentrations calculated based on a standard curve of purified human TAT complex, read on an Envision 2105 Multimode Plate Reader (2105-0010, Perkin Elmer). Concentration data were plotted, and statistics were calculated using GraphPad Prism v8.2.0.

SIRPα reporter assay

The PathHunter Jurkat SIRPα Signaling Bioassay Kit (93-1135Y19, Eurofins DiscoverX) was used to assess human CD47 function. Porcine KECs (30,000 cells per well) were incubated with or without increasing concentrations of an anti-human CD47 blocking antibody (clone B6H12.2) followed by addition of Jurkat SIRPα reporter cells (10,000 cells per well). Cells were incubated at 37 °C in a humidified CO2 incubator (5% CO2) for 24 h, and kit instructions were followed for signal detection. Luminescence of the plates was read on a microplate reader (FilterMax F5, Molecular Devices) according to the manufacturer’s instructions. Relative luminescence units were plotted and statistics were calculated using GraphPad Prism v8.2.0.

Caspase 3/7 assay

Endothelial cells (70% confluent in a 10-cm gelatin-coated tissue culture dish) were treated with 50 ng ml−1 recombinant human TNF (210-TA-100/CF, R&D Systems) in complete endothelial cell medium overnight. Cells were harvested from the plate and 10,000 cells were added to a 96-well flat-bottom white plate (07200589, Thermo Fisher), and caspase 3/7 activity was determined using the Caspase-Glo 3/7 Assay System (G8093, Promega). In brief, an equal volume of fresh Caspase-Glo 3/7 reagent was added to the wells, the plate incubated for 30 min at room temperature in the dark and then read on the Envision 2105 Multimode Plate Reader (2105-0010, Perkin Elmer). Data were plotted, and statistics were calculated using GraphPad Prism v8.2.0.

NHP recipients

Male and female cynomolgus monkeys (Macaca fascicularis; Bioculture US LLC and Alpha Genesis) weighing 4–12 kg (estimated 3–8 years of age) were used. Monkeys were screened for the presence of anti-porcine IgG and IgM (described above), and animals with low anti-porcine IgG and anti-IgM were selected as recipients. Yucatan pigs weighing 5–27 kg were used as kidney donors. All animal care, surgical procedures and postoperative care of animals were conducted in accordance with NIH Guidelines for the care and use of primates and The Guide for the Care and Use of Laboratory Animals and were approved by Institutional Animal Care and Use Committees at Duke University (protocol A032-20-02, approved 27 February 2020), University of Wisconsin at Madison (protocol G006507, approved 30 September 2021) and the Massachusetts General Hospital (protocol 2017N000216, approved 20 November 2020). All studies followed the 3Rs principles.

Kidney transplantation

To prepare the cynomolgus monkey for the procedure, a central venous line was inserted through the internal jugular vein 2–7 days before kidney transplantation. Through the midline incision, the kidney xenograft was transplanted intraperitoneally by anastomosing the renal vein and artery to the vena cava and abdominal aorta, respectively. Ureterovesical anastomosis was performed by the Lich–Gregoir technique without placing a ureteral stent. Bilateral native nephrectomy was performed simultaneously in the majority of the recipients with the exception of M8220 and M6521, where one native kidney was left intact at transplant and then removed around POD 20. Postoperatively, the transplanted kidney was monitored by urine output, ultrasound, clinical chemistry and haematology, as well as protocol biopsies. The central venous line was removed by 2–4 weeks once recipient animals had stable kidney function, to avoid the risk of infection. Long-term survival refers to life-supporting function of more than 3 months in the NHP recipient.

Histopathological analysis

Protocol renal biopsies were obtained every 2–4 months in recipients with stable function as well as when rejection was suspected owing to a rise in creatinine. Tissue was processed for light microscopy. Following euthanasia of a monkey, a complete autopsy was performed for histopathological examination of the renal xenograft, lymph nodes, heart, lung, liver, pancreas, thymus and skin. Xenograft haemotoxin and eosin and periodic acid–Schiff-stained samples were scored by current Banff criteria including C4d deposition42.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.