iPSC differentiation and QC: why the same protocol varies

A protocol can fix the order of development

Differentiation retraces, in a dish, the order in which cells form in the body.

You cannot turn an iPS cell straight into heart or nerve. The cells pass through intermediate stages, such as mesoderm or ectoderm, and each stage gets a different agent, or signal. For cardiomyocytes, you first activate the Wnt pathway with a GSK3 inhibitor such as CHIR99021, and several days later you switch the Wnt pathway off. For neurons, inhibiting the SMAD pathway with two inhibitors at once brings neural progenitor markers up in most cells in about 11 days. Retinal pigment epithelium is slower: expect 40 to 60 days before pigmented cells line up.

All of this can go into a protocol, because research has settled which agents go in and in what order. What research has not settled for you is how exactly the concentration and the switching time act on the cells in your hands.

A protocol decides the agents and their order, while concentration and timing are the part you re-verify for each cell line.

Why does the same protocol give different results in different lines?

Apply one protocol to a different line, and efficiency can drop.

Studies have shown that lines differ in how readily they differentiate. The Wnt switch for cardiomyocytes is a step where yield swings widely if concentration or timing is off, so the working assumption is to re-titrate for every line. The starting state matters as well. If the medium or matrix drifts while you maintain the undifferentiated state, or if repeated passaging changes the cells, they do not respond uniformly when differentiation begins. That is why maintenance is fixed with a feeder-free matrix and a medium of defined composition. Matrices have moved from basement membrane extracts such as Corning's Matrigel to ones of known composition, such as the laminin-511 E8 fragment. Among media, Ajinomoto's StemFit, STEMCELL Technologies' mTeSR and Thermo Fisher Scientific's Essential 8 are the leading names.

Commercial kits exist for the differentiation stage too, in differentiation media and induction reagents, such as STEMCELL Technologies' STEMdiff. The efficiency in a kit's instructions, though, is the result in the line used for evaluation.

Start optimizing conditions only after the line is chosen. If you lock conditions before the master cell bank is made, you redo the work the moment the line changes. If you are in sales or purchasing, ask which line the data came from when you propose or are offered a kit or medium. The answer tells you whether the same result is likely in your line.

When the line changes, condition finding and evaluation data go back to the start, so choosing the line first costs the least rework.

Can you purify the target cells inside the process? Lactate and cardiomyocytes

Differentiation alone does not give you 100% of the cell you want. You need a step that raises the share of target cells and removes those that did not differentiate fully.

Metabolic selection of cardiomyocytes is an elegant case. Undifferentiated iPS cells live on glucose, but cardiomyocytes can burn lactate. Remove the glucose and supply only lactate, and cells other than cardiomyocytes cannot survive, which brings cardiomyocyte purity to nearly 99% (Tohyama et al., 2013). Unlike selection by surface markers, it needs no antibody and no cell sorter.

How you purify also shapes how you design the differentiation. A cell type that lactate can select can be cleaned inside the process, which lightens the testing downstream. Where no such metabolic difference exists, you must select by marker or raise the differentiation efficiency itself.

Whether purification can be built into the process decides how heavy your later burden of explaining quality will be.

Residual undifferentiated cells: detection limits differ by orders of magnitude

After purification comes the question of how you know the unwanted cells are gone.

Undifferentiated iPS cells left in the product could form tumors after transplant. So you must show that the level is "not zero, but too low to detect," and the detection method decides how convincing that is. To see tumorigenicity directly, there is transplantation into immunodeficient animals. Other options are colony formation in soft agar, flow cytometry for TRA-1-60 or SSEA-4, and qRT-PCR for undifferentiated marker genes. A list of detection methods is in the directory.

By orders of magnitude, qRT-PCR is the most sensitive. A system targeting LIN28 was reported to detect down to 0.002% (Kuroda et al., 2012). High sensitivity is not the same as measuring correctly, though. The detection limit moves with what you define as an undifferentiated marker, and if differentiated cells also express a little of that gene, the apparent sensitivity drops.

Managing residual undifferentiated cells means choosing a method and being able to explain that method's detection limit.

Genome changes and allogeneic stocks: fewer lines, more depth

Long culture changes the genome too, because cells carrying a mutation that favors culture are selected and expand.

A real case is the first clinical study of autologous iPS-derived retinal pigment epithelium. The first patient went on to transplant. In the second, copy number changes were found in the cells and the transplant was dropped (Mandai et al., 2017). Genome testing, in other words, has stopped a product. What karyotyping and digital PCR can and cannot see is covered in karyotyping and genomic stability of iPS cells (in Japanese). For how yield and quality move when differentiation is scaled up into large vessels, see our article on a paper that expanded iPS cell-derived islets fivefold in suspension culture (in Japanese).

A second way to reduce variation is an allogeneic stock. With autologous cells every patient has a different line, so condition finding and testing pile up by the number of patients. Allogeneic production makes doses for many patients from lines established in advance. The CiRA Foundation at Kyoto University established 27 lines from 7 donors who are HLA homozygous, meaning they inherited the same HLA type from both parents, and estimates that the lines match about 40% of Japanese people. The same estimate puts coverage at about 50%, 80% and 90% if the stock grows to 10, 75 and 140 HLA types.

Fewer lines make it worthwhile to work out each line's differentiation conditions and characterization thoroughly. The comparison of the two approaches is in autologous and allogeneic cell therapy (in Japanese).

Autologous production spreads quality assurance across as many lines as there are patients, while allogeneic production concentrates it on a few.

Five checks, in the order that avoids rework

Whether you work in development, manufacturing, quality assurance or purchasing, checking an iPSC-derived product in this order reduces rework.

1. Is the line decided? Is it autologous or allogeneic, and if allogeneic, which line from which stock?
2. For that line, is there a record of re-titrating the differentiation concentration and timing?
3. By which indicator, and at what point, is the share of target cells measured?
4. By which method are residual undifferentiated cells detected, and what detection limit in percent do you claim?
5. At which passage is the genome examined, and by karyotyping or digital PCR?

The point where the process switches from keeping cells undifferentiated to removing undifferentiated cells is what shapes how you explain quality. For product comparisons, see the directory pages for undifferentiated-state media, differentiation media, induction and competence reagents and karyotyping. Where this step sits in the whole iPSC process is covered on our iPS cell process map, and the full picture of quality tests for cell therapy is in critical quality attributes for cell therapy (in Japanese).