Failure to replicate the STAP cell phenomenon

arising fromH. Obokata et al. Nature 505, 641–647 (2014) doi:10.1038/nature12968; retraction 511, 112 (2014) doi:10.1038/nature13598 ; and H. Obokata et al. Nature 505, 676–680 (2014) doi:10.1038/nature12969; retraction 511, 112 (2014) doi:10.1038/nature13599

Although the reports that stress (such as exposure to acid) can coax somatic cells into a novel state of pluripotency^1,2 have been retracted^3,4, the validity of stimulus-triggered acquisition of pluripotency (STAP) remains unclear (http://dx.doi.org/10.1038/protex.2014.008 and Supplementary Information). Here we describe the efforts of seven laboratories to replicate STAP, including experiments performed within the laboratory where STAP first originated, as well as re-analysis of the sequencing data from the STAP reports. Neonatal cells treated with two STAP protocols exhibited artefactual autofluoresence rather than bona fide reactivation of an Oct4 (also known as Pou5f1) and green fluorescent protein (GFP) transgene reporter, did not reactivate pluripotency markers towards embryonic stem (ES)-cell-like levels, and failed to generate teratomas or chimaerize blastocysts. Re-analysis of the original RNA sequencing (RNA-seq) and chromatin immunoprecipitation sequencing (ChIP-seq) data identified discrepancies in the sex and genetic composition of parental donor cells and converted stem cells, and revealed a STAP-derived cell line to be a mixture containing trophoblast stem cells, attesting to the importance of validating the properties and provenance of pluripotent stem cells using a wide range of criteria.

To assess the reprogramming capacity of STAP protocols, we used a transgenic Oct4-GFP reporter, which shows GFP reactivation during Oct4/Sox2/Klf4 reprogramming, in established induced pluripotent stem (iPS) cells and in the gonads of mid-gestation ‘all iPS cell’ embryos generated by tetraploid complementation^5,6,7 (Extended Data Figs 1 and 2a). Working within the Vacanti laboratory where the concept of STAP cells originated, and assisted by a co-author of the STAP papers, a Daley laboratory member (A.D.L.A.) attempted to replicate two reported STAP protocols: (1) mechanical trituration and acid treatment of mouse lung cells (Brigham and Women’s Hospital (BWH) protocol; see Supplementary Information), and (2) acid treatment of mouse splenocytes (RIKEN protocol; Methods and Extended Data Fig. 2b). Seventy-two hours after stress treatment of lung cells, floating spheres appeared amidst cellular debris. Fluorescence microscopy revealed that both Oct4-GFP and wild-type spheres emitted low-level broad spectrum fluorescence detectable within both green and red filters, indicating autofluorescence (Fig. 1a). Untreated Oct4-GFP ES cells did not emit the same low-level broad spectrum fluorescence as STAP-treated cells. STAP-treated splenocytes formed spheres with lower efficiency, but also appeared autofluorescent.

Figure 1: Characterization of the STAP cell phenomenon.

See Supplementary Information for further details. a, STAP treatment produces fluorescent signal detected in both FITC (green) and TRITC (red) channels in STAP-treated Oct4-GFP and wild-type (WT) cells, consistent with autofluorescence. TRITC signal was not detected in control Oct4-GFP mouse ES cells. Note saturation of the green signal in Oct4-GFP ES cells at the higher exposure time required to detect FITC from autofluorescent spheres. b, Absence of ES-cell-like Oct4-GFP reactivation. Representative flow cytometry results 7 days after STAP treatment of lung cells or splenocytes (BWH or RIKEN protocol, respectively) without singlet/doublet exclusion and live/dead-cell discrimination. GFP gates were calibrated based on control Oct4-GFP ES cells grown on feeders. Whereas control ES cells are bright and situated at approximately 1 × 10⁵ (arbitrary units), no event resembling Oct4-GFP ES cells was detected after STAP treatment. One replicate per protocol is shown. iMEFs, irradiated MEFs. c, STAP-treated cells do not form teratomas using PLGA-based teratoma production methods¹. Photograph of control mouse ES-cell-derived teratoma (top left) and non-teratoma STAP-PLGA mass (bottom left). Representative haematoxylin and eosin (H&E) stainings of a control mouse ES-cell-derived teratoma (top right) and the non-teratoma STAP-PLGA mass (bottom right). d, STAP-PLGA mixtures present no indication of ES-cell-like in vivo differentiation capacity after injection into immunocompromised mice. Note lack of organization into representative tissue structures typically observed in ES-cell-derived teratomas. DAPI, 4′,6-diamidino-2-phenylindole. e, STAP-treated lung cells fail to incorporate into preimplantation embryos after morula aggregation. f, Analysis of sequencing data. Samples are classified based on copy number and genotype. STAP cells, STAP stem cells (STAP-SCs) and ES cells share similar characteristics for genotype and copy number of chromosome X. g, Copy number (CN) profiles, reported as a log₂ ratio (observed to expected read counts), derived using ChIP-seq input data. Red/green correspond to significant amplifications and deletions (log₂(CN) ≥ 0.2 or ≤ −0.2 and P ≤ 0.01), respectively. Grey denotes non-significant variants. Note the amplifications of chromosomes 8 (FI-SCs) and 6/11 (TSCs) and the single copy of chromosome X in STAP cells, STAP-SCs, FI-SCs and ES cells. h, SNVs inferred from RNA-seq data using the mouse reference genome (derived from C57BL6 strain). The selected SNVs are classified as homozygous for reference allele (0/0 genotype), homozygous for alternative allele (1/1 genotype) or heterozygous (0/1 genotype). Samples are clustered based on the sum of edit distance between each SNV. Note that each pair of replicates is always grouped together. A subset of samples (CD45⁺, STAP, STAP-SCs and ES cells) shows prevalence of heterozygous alleles (A); FI-SC samples have prevalence of homozygous alleles for the reference variant (B); and, TSC and epiblast stem cell (EpiSC) samples have a larger number of homozygous alternative alleles (C). i, Contamination in the FI-SC samples with TSCs. The expected frequency of reads covering the alternative allele for heterozygous SNVs is ∼50%, which is observed in all samples including TSCs (left). In FI-SCs, it was ∼12% (Extended Data Fig. 3), suggesting false-positive calls or contamination. The alternative allele frequency distributions of TSC homozygous and heterozygous SNVs sets in FI-SCs (right) show peaks at 9% and 4%, respectively. These results indicate that FI-SC samples are approximately 10% contaminated by TSC samples. Original magnifications, ×20 (a, d, e) and ×4 (c).

PowerPoint slide

Flow cytometry indicated STAP-treated Oct4-GFP cells did not exhibit Oct4-GFP reactivation at levels comparable to control Oct4-GFP mouse ES cells, and were indistinguishable from stressed wild-type controls (Fig. 1b). Absence of ES-cell-like levels of Oct4, Sox2 and Nanog transcripts and nonspecific immunofluorescence corroborated flow cytometry data (Extended Data Fig. 2c, d). Rare pluripotent cells should generate teratomas in immunocompromised mice^8,9, but STAP cells could not, unlike control ES cells (Extended Data Fig. 2e, f). Replication of the poly-l-glycolic acid (PLGA)-based teratoma production method described in the original STAP reports with GFP cells to distinguish host and donor contribution produced distinct masses of connective tissue, muscle and scar, with minimal GFP content, indicating primarily host origin (Fig. 1c, d and Extended Data Fig. 2g). Rare GFP-positive clusters did not form differentiated tissues characteristic of ES-cell-derived teratomas (Fig. 1d). Autofluorescent spheres failed to enter development after morula aggregation or blastocyst injection (Fig. 1e and Extended Data Fig. 2h–j). Therefore, pluripotency was undetectable in STAP experiments. Six other laboratories (Deng, Hanna, Hochedlinger, Jaenisch, Pei and Wernig) also attempted to generate STAP cells (Table 1) and made the following observations. First, autofluorescent sphere-like aggregates after STAP treatment were universally seen. Second, transgenic reporters used by Obokata and colleagues (GOF18-Oct4-GFP, containing the 18-kilobase genomic Oct4 fragment (GOF18)) and by the Daley, Pei and Hanna laboratories (GOF18-Oct4ΔPE-GFP, lacking the Oct4 proximal enhancer (PE) element) both exhibit activity in pre-implantation embryos, early post-implantation epiblast cells (embryonic day (E) 5.5), germ cells, and mouse ES/iPS cells; however, differential activity in late post-implantation epiblast (E6.5) and early passage mouse epiblast-derived stem cells has been ascribed to the Oct4 proximal enhancer^10,11,12. Using the same reporter as Obokata and colleagues^1,2, the Deng laboratory observed that the GFP signal in chemical iPS cells was easily distinguishable from the autofluorescence of STAP-treated cells (Extended Data Fig. 2k). The Jaenisch, Wernig and Hochedlinger laboratories failed to observe GFP reactivation with Oct4 or Nanog knock-in reporters, excluding a scenario of uncoupling between GFP and endogenous pluripotency expression¹⁰. Despite a range of tested reporters, no group documented authentic Oct4/Nanog reporter activation that resembled bona fide ES cells. Third, the Deng laboratory failed to observe Oct4, Sox2 and Nanog induction 3 and 7 days after STAP treatment, reducing the likelihood that pluripotency was transiently activated and silenced by day 7 (Extended Data Fig. 2l). Finally, the Hanna, Wernig and Hochedlinger laboratories failed to generate stem-cell lines by culturing STAP-treated cells in leukaemia inhibitory factor (LIF) and adrenocorticotropic hormone (ACTH)-supplemented medium. In summary, 133 replicate attempts failed to document generation of ES-cell-like cells, corroborating and extending a recent report¹³.

Table 1 Global efforts to replicate STAP cell generation

We re-examined the high-throughput sequencing data from the STAP reports to investigate the genetic provenance of parental CD45⁺ cells and converted STAP cells, STAP stem cells and Fgf4-induced stem cells (FI-SCs) (Fig. 1f). Comparative genomic hybridization array data mentioned in the original paper¹ were not publicly released. Copy number variation (CNV) analysis conducted using ChIP-seq input samples revealed a discrepancy in sex across samples as well as chromosomal aberrations (Fig. 1g). In the original STAP reports, the authors stated that they mixed CD45⁺ cells from male and female mice owing to the small number of CD45⁺ cells retrieved from individual neonatal spleens. However, our analysis indicates that CD45⁺ cells were female, whereas the derived cells (STAP cells, STAP stem cells and FI-SCs) were all male, a clear inconsistency. We note that control ES cells were also male (Fig. 1g). FI-SCs possessed trisomy 8, which renders mouse ES cells germline-incompetent¹⁴ (Fig. 1g).

Inferred single nucleotide variants (SNVs) from RNA-seq data allowed classification of samples as genetically similar or dissimilar (Fig. 1h). Control ES cells, parental donor female CD45⁺ cells, STAP cells, and STAP stem cells all possessed similar SNV profiles, consistent with their derivation from a first generation hybrid of C57BL6/129 strains, the reported genotype (Fig. 1h and Extended Data Fig. 3). By contrast, FI-SCs had an SNV profile that matched a single nucleotide polymorphism (SNP) profile of C57BL6 strain origin, indicating distinct genetic provenance from parental CD45⁺ and STAP samples (Fig. 1h and Extended Data Fig. 3). Independently sourced control epiblast stem cells and trophoblast stem cells (TSCs) had SNV profiles divergent from the CD45⁺ and STAP sample cohort, as expected (Fig. 1h). An anomalous allele frequency distribution observed in FI-SCs, and reciprocal analyses of FI-SC heterozygous SNVs in TSCs and TSC homozygous and heterozygous SNVs in FI-SCs, revealed that FI-SCs were derived from a C57BL6 strain origin, with approximately 10% contamination from TSCs (Fig. 1i and Extended Data Fig. 3). These are concordant with the findings from a recent RIKEN report (http://www3.riken.jp/stap/e/c13document52.pdf). This contamination with TSCs explains the high-grade placenta-forming capacity reported for the Fl-SCs², an unusual feature that implied totipotency, but which seems to have been due to admixture of cells.

In summary, our replication attempts and genetic analysis indicate that existing STAP protocols are neither robust nor reproducible. To substantiate future claims of reprogramming and alternative states of potency, we urge a rigorous application of several independent means for validating functional pluripotency and genomic profiling to confirm cell line provenance. Ultimately, the essential standard of robustness and reproducibility must be met for new claims to exert a positive and lasting influence on the research community.