A comparison of non-integrating reprogramming methods

Human induced pluripotent stem cells (hiPSCs) are useful in disease modeling and drug discovery, and they promise to provide a new generation of cell-based therapeutics. To date there has been no systematic evaluation of the most widely used techniques for generating integration-free hiPSCs. Here we compare Sendai-viral (SeV), episomal (Epi) and mRNA transfection mRNA methods using a number of criteria. All methods generated high-quality hiPSCs, but significant differences existed in aneuploidy rates, reprogramming efficiency, reliability and workload. We discuss the advantages and shortcomings of each approach, and present and review the results of a survey of a large number of human reprogramming laboratories on their independent experiences and preferences. Our analysis provides a valuable resource to inform the use of specific reprogramming methods for different laboratories and different applications, including clinical translation.

A n A ly s i s Human induced pluripotent stem cells (hiPSCs [1][2][3] ) are useful in disease modeling and drug discovery, and they promise to provide a new generation of cell-based therapeutics. To date there has been no systematic evaluation of the most widely used techniques for generating integration-free hiPSCs. Here we compare Sendai-viral (SeV) 4 , episomal (Epi) 5 and mRNA transfection mRNA 6 methods using a number of criteria. All methods generated high-quality hiPSCs, but significant differences existed in aneuploidy rates, reprogramming efficiency, reliability and workload. We discuss the advantages and shortcomings of each approach, and present and review the results of a survey of a large number of human reprogramming laboratories on their independent experiences and preferences. Our analysis provides a valuable resource to inform the use of specific reprogramming methods for different laboratories and different applications, including clinical translation.
The goal of this analysis was to compare non-integrating reprogramming methods as they are being practiced now, using readily available and widely used reagents and kits. In SeV reprogramming 4 , Sendaiviral particles are used to transduce target cells with replicationcompetent RNAs that encode the original set of reprogramming factors (OCT4, SOX2, KLF4 and cMYC (together referred to as OSKM)). Here we used the Cytotune kit (Life Technologies). In Epi reprogramming 5 , prolonged reprogramming factor expression is achieved by Epstein-Barr virus-derived sequences that facilitate episomal plasmid DNA replication in dividing cells. Human episomal reprogramming was first realized by the Thomson laboratory 7 ; here we use a more efficient method that employs the reprogramming factors OCT4, SOX2, KLF4, LMYC and LIN28A combined with P53 knock-down (shP53) 5 .
In mRNA reprogramming 6 , cells are transfected with in vitrotranscribed mRNAs that encode OSKM, the additional reprogramming factor LIN28A and GFP. Several chemical measures are employed to limit activation of the innate immune system by foreign nucleic acids 6 , and, due to the very short half-life of mRNAs, daily transfections are required to induce hiPSCs. Here we used the mRNA reprogramming kit from Stemgent.
First, we looked at reprogramming efficiencies. Determining precise reprogramming efficiencies is complicated by differences between protocols in cell passaging regimes, cell plating efficiencies, as well as by variable somatic cell proliferation and transfection/transduction rates; nevertheless, the number of hiPSC colonies generated per somatic input cell is an important parameter. Of the three non-integrating methods, mRNA-based reprogramming was found to be the most efficient (Fig. 1a). The mean efficiency of successful reprogramming experiments was 2.1% for the mRNA method (n = 3 successfully reprogrammed samples), followed by SeV (0.077%) and Epi (0.013%) reprogramming; the differences in efficiencies (mRNA vs. Epi, mRNA vs. SeV, Epi vs. SeV) reached statistical significance (P < 0.05, Student's t-test). For comparison, lentiviral (Lenti) reprogramming (with the reprogramming factors OSKM) generated colonies with an efficiency of 0.27% (n = 7). Efficiencies can be sample-dependent; however, the subset of samples that were successfully reprogrammed by all four methods (one neonatal (BJ) and two patient-derived lines (PS1, PS2)) showed the same trend and rank order (gray bars in Fig. 1a). Furthermore, our results are consistent with those reported by others 5,6,[8][9][10][11] (black bars in Fig. 1a).
Next, we considered the success rates, defined as the percentage of samples for which at least three hiPSC colonies emerged (Fig. 1b). In our hands, the Lenti (100% success rate), Epi (93%) and SeV (94%) methods very reliably generated multiple hiPSC colonies.
A comparison of non-integrating reprogramming methods A n A ly s i s In contrast, with the mRNA method, the success rate was significantly lower (27%, P < 0.001, Fisher's exact test). Failures did not appear to be due to reduced mRNA transfection efficiencies (GFP expression); rather, they were associated with massive cell death and detachment. Furthermore, whereas skin fibroblast samples BJ, PS1 and PS2 were readily reprogrammed using all methods, two other patient skin samples (PS3, PS4) that could be reprogrammed using Epi and SeV methods failed with the mRNA method, strongly suggesting that these failures were method-specific and sample-dependent. When we used a modified protocol that employed transfection of microRNAs (miRNAs) (miRNA Booster Kit, Stemgent) and mRNAs, the success rate improved significantly, to 73% overall (P < 0.05) and to 100% for samples refractory to reprogramming by mRNA alone (n = 4). The mean reprogramming efficiency of miRNA + mRNA reprogramming was 0.19% for the 11 fibroblast samples that were reprogrammable with this method.
To allow us to directly compare the workload of generating hiPSCs with the three non-integrating methods, we measured the hands-on time required, including reagent, media and feeder cell preparations, from initial seeding of the target somatic cells to the picking of hiPSC colonies (Fig. 1c,d). The SeV method demanded the least amount of work, consuming 3.5 h of hands-on time until colonies were ready for picking around day 26. Epi reprogramming consumed about 4 h, with colonies large enough for picking appearing around day 20, and the miRNA + mRNA method required about 8 h, although colonies were ready to be picked around day 14. SeV and Epi reprogramming required a larger starting cell number and that more clones be expanded and tested for the loss of the reprogramming agents (see below), adding to the workload (Fig. 1c).
Several studies have reported small genomic alterations in hiPSC lines, including copy number variations (CNVs) and coding mutations, possibly reflecting a mutagenic effect of the reprogramming process itself 12,13 . We conducted a small-scale genome-wide array comparative genomic hybridization (aCGH) analysis of representative patient-derived Lenti, SeV, Epi and RNA hiPSC lines and found the majority of CNVs preexisting and the frequency of possible de novo aberrations uniformly low across all methods (Supplementary Fig. 1b).
Next, we tested how quickly the exogenous reprogramming agents were lost during hiPSC expansion. Because modified mRNAs have a short half-life and cannot integrate or replicate, we focused on SeV and Epi hiPSC samples. Loss of SeV RNA was independent of sample type except for erythroblast-derived SeV hiPSCs, which showed a somewhat delayed loss of SeV RNA (Fig. 2b). We observed a passage-dependent decrease in the number of hiPSC lines retaining SeV RNA, from 100% at passage 1-5, to around 53.8% at passage 6-8 (66.7% including erythroblast hiPSC samples) and down to 21.2% (34.3% including erythroblast hiPSC samples) at passage 9-11. EBNA1 DNA was detected in ~39.1% of Epi hiPSCs analyzed before passage 6. This rate decreased, but slowly, to 37.9% at passages 6-8 and 33.3% at passage 9-11 (Fig. 2c), a frequency identical to the one observed by Okita et al. 5 . Most lines either showed an increase or a decrease in EBNA1 DNA levels, resulting in a distinct separation into EBNA1 DNA negative and EBNA1 DNA high lines by passage 10. PCR analyses revealed that O4-shP53 plasmid sequences were retained in 13/14 higher-passage DNA high lines (Supplementary Fig. 2b). A few of the O4-shP53 plasmid-positive clones also contained the LIN28A-LMYC plasmid, whereas sequences derived from the SOX2-KLF4 encoding were not detected in this sample set. Retention of episomal plasmid sequences in rare hiPSCs that may have a growth advantage is a potential concern that necessitates continued vigilance. To address this issue, we developed a fluorescent protein (H2B-mKO2)-tagged npg A n A ly s i s version of the OCT4-shP53 episomal plasmid that facilitates identification of plasmid-retaining colonies (Fig. 2d).
When we extended our analysis to genes previously reported as differentially expressed between hESC and hiPSC lines (TCERG1L, FAM19A5 (refs. 14,15) and MEG3/RIAN 16 ), we found that they were indeed differentially expressed, but only in some of the hiPSC lines. Notably, all reprogramming methods performed quite similarly, with a small fraction of all method-specific sets of the lines showing the 'correct' hESC-like gene expression levels for the three variable genes (Supplementary Fig. 3). Our observations are in agreement with reports showing that aberrant methylation of the TCERG1L gene locus 17 or aberrant expression of Meg3/Meg8 (ref. 16) is observed only in some iPSC lines.

A n A ly s i s
To evaluate their pluripotency, we differentiated Lenti, Retro, as well as reprogramming factor-free SeV, mRNA and Epi hiPSCs derived from two samples (BJ, PS1) as embryoid bodies and determined each cell line's differentiation propensity for the three germ layers using the scorecard approach 18 . All lines scored as pluripotent, with differentiation score averages well within the non-outlier range of the human pluripotent reference set and with no apparent trends specific to the reprogramming method (Fig. 2h).
Finally, to allow us to compare our observations with those of other human cell reprogramming laboratories, we designed an online survey (Supplementary Fig. 5). The survey was distributed to a set of hiPSC core facilities and research laboratories. We received 55 responses from laboratories representing 12 countries and an estimated combined experience of >1,450 human somatic cell reprogramming experiments. The most commonly attempted methods for human fibroblast reprogramming were Lenti (38 laboratories), Retro (33) and SeV (35), followed by RNA (22) and Epi (21) methods ( Table 1). All laboratories were able to successfully use all these methods on fibroblasts, with the exception of RNA reprogramming, which could not be established in 41% of laboratories. When we focused on just the responses of the most experienced laboratories (>20 reprogramming samples processed), a similarly high fraction (47%) reported an inability to generate RNA-hiPSCs, suggesting that the difficulties associated with the RNA method are not merely due to a lack of sufficient reprogramming expertise. The survey responses allowed us to calculate method 'adoption' and 'rejection' rates specifically for those laboratories that were able to establish the method. Very high methodadoption rates were reported for SeV, whereas the rates for Retro and/or Lenti methods were low, indicating that most laboratories eventually switched to non-integrating techniques. Most of the reported experience with fibroblast reprogramming extended to blood cells, with a few notable exceptions. None of the laboratories reported successful RNA reprogramming of hematopoietic cells, and although Retro reprogramming of blood cells was seldom if ever practiced, Lenti reprogramming was more popular with blood cells than with fibroblasts.
In brief, we detected no substantial method-specific differences in marker expression levels or patterns, developmental potential, DNA methylation or CNV loads of euploid lines. It is possible that more detailed studies will reveal method-specific differences, such as in the potential to form specific cell types. For example, the potential of miPSCs was recently shown to depend on the factors used for reprogramming 19 . Nevertheless, our conclusion that no overt method-specific differences exist echoes findings by other groups who performed less comprehensive cross-comparisons of smaller sets of reprogramming methods 8,12 .
SeV reprogramming was efficient and highly reliable, with a low workload and a complete absence of viral sequences in most lines at higher passages. The shortcomings of SeV reprogramming include the dependence on one commercial vendor (making it, for example, difficult Experienced refers to laboratories that have successfully reprogrammed at least 20 human cell samples; Tried to the number of laboratories that reported having attempted to reprogram somatic cells using the indicated method (RNA refers to the mRNA method with or without miRNA use); Success to the number and percentage of these laboratories that were able to derive hiPSC lines with the method. The survey also included questions about the method-specific usage frequencies. Rejected refers to the % of successful laboratories reporting usage frequencies of ≤10% (≤10% of samples reprogrammed using this method); Adopted refers to usage frequencies of ≥60%. The survey questionnaire is provided as Supplementary Figure 5. The main advantages of the RNA method are the speed of colony emergence, high efficiency, a complete absence of integration, a very low aneuploidy rate and a low donor cell requirement (typically 50,000 cells, but as few as 1,000 human fibroblasts can be reprogrammed; data not shown). Minor issues are the increased workload and the need for a tissue culture incubator with O 2 control. More significantly, RNA reprogramming was successful only with a small fraction of the samples; we and others 8 frequently experienced difficulties when attempting to reprogram primary fibroblasts obtained from patient skin biopsies. Problems with implementing this technique were also reported in our survey, and to date there have been no reports of successful generation of RNA-hiPSCs from blood cells. One promising approach for improving RNA-based reprogramming is the inclusion of pluripotency-inducing miRNAs (Fig. 1b). Others have reported increased RNA reprogramming efficiencies and accelerated colony emergence by fusing Oct4 to a heterologous transactivation domain 10 . The conventional RNA method can therefore be useful for easy-toreprogram fibroblast samples, but needs to be further optimized to overcome the reprogramming resistance and excessive cell death observed with many patient samples.
Key advantages of Epi reprogramming are the high reliability of hiPSC generation from fibroblast and blood samples (CD34 + and peripheral blood mononuclear cells (PBMNCs)) 20,21 ; see also Supplementary Figs. 6 and 7) and the quick loss of the reprogramming agent (relative to SeV). However, Epi reprogramming does raise concerns regarding the genetic integrity of the resulting hiPSC lines due to the use of a TP53 short hairpin RNA (shRNA) cassette. Notably, we did not observe an increase in CNVs, and the rate of aneuploidies, although elevated relative to SeV, RNA and Lenti hiPSCs, was still lower than what we observed with Retro hiPSCs. To further boost Epi reprogramming efficiencies, several groups have successfully employed small molecules 22,23 , increased levels of EBNA1 (ref. 21), or used additional or modified reprogramming factors (such as BCL-XL 24 or OCT4-VP16 (ref. 25)).
Although xeno-free procedures have been described for all major reprogramming methods 8,10,23,26-28 , Epi reprogramming seems particularly well-suited for clinical translation because it is integrationfree, works reliably with patient fibroblasts and blood cells, and is based on a very simple reagent (plasmid DNA) that can easily be generated using current good manufacturing practice (cGMP)compatible processes. SeV, on the other hand, is currently not available commercially as a cGMP-grade reagent for reprogramming, although Sendai virus has been used as a vehicle in human vaccination studies. Another important consideration are the restrictions imposed by the respective licensors or vendors. In general, the commercial kits are restricted to educational use and noncommercial or nonclinical in vitro research; however, additional rights to commercial or clinical use can be acquired (Supplementary Table 1).
In summary, the choice of reprogramming method will depend on each laboratory's particular requirements ( Table 2). The field of pluripotency induction continues to develop at a rapid pace, as evidenced by the recent publication of a gene-free, small moleculebased method 29 , as well as the identification of novel pathways that can be manipulated to augment the efficiency and completeness of reprogramming 30 . The focus of technology development efforts should now move toward leveraging these findings to generate improved methods for clinical translation of human pluripotent stem cells.

METHOdS
Methods and any associated references are available in the online version of the paper.
Note: Any Supplementary Information and Source Data files are available in the online version of the paper.