Induction of neuronal cells from liver cells
To test whether cells derived from liver can be induced to become neuronal cells, we established primary liver cultures from postnatal days (P) 2-5 wild type and TauEGFP knock-in mice (Tucker et al., 2001
; Wernig et al., 2002
). Four days after isolation, the majority of cells showed a typical epithelial morphology and expressed albumin, α-fetoprotein and α-anti-trypsin ( and S1A
). One week post explantation, a typical culture was composed of 60% albumin-positive hepatocytes, 16% myeloid cells, 2% Kupffer cells and 2% endothelial cells (Figure S1B,D
). Absence of the neuronal or neural progenitor cell markers Sox2, Brn2, MAP2, and NeuN in the culture was confirmed by immunofluorescence (data not shown). The rare (~1/5000) Tuj1-positive cells had a flat morphology, not resembling neuronal cells (Figure S1C
). TauEGFP-positive cells were not detectable in these cultures as evaluated by flow cytometry or fluorescence microscopy.
Induction of neuronal cells from liver cells
The primary liver cultures were then replated and infected with doxycycline (dox)-inducible lentiviruses containing the cDNAs of Ascl1 (A), Brn2 (B) and Myt1l (M) in various combinations. Thirteen days after addition of dox, TauEGFP-positive cells with a complex neuronal morphology were readily detected in the wells that received all three factors (BAM) (). No neuronal cells were found in any other combination (). Immunofluorescence confirmed that all TauEGFP-positive cells generated by the BAM factors were also Tuj1-positive (). When analyzed three weeks after infection, the cells expressed the additional pan-neuronal markers PSA-NCAM, NeuN, MAP2, and synapsin (). A fraction (35 out of 200 counted Tuj1-positive cells) of the cells could also be labeled with an antibody against vesicular glutamate transporter 1 (vGLUT1) (). In contrast, no GAD67-, TH-, ChAT-, or serotonin-positive cells were detected (0 out of at least 200 counted Tuj1-positive cells). As neuronal subtype-specific markers are expressed predominantly in mature stages of neuronal differentiation while Tuj1 labels already early postmitotic immature neurons we conclude that the majority of mature iN cells are excitatory neurons. Moreover, qRT-PCR analysis showed that TauEGFP-positive iN cells 3 weeks after infection had not only induced neuronal transcripts but efficiently silenced transcripts characteristic of the starting cell population ().
Genetic proof that iN cells can be derived from albumin-expressing hepatocytes
We then employed the Cre-LoxP system to unambiguously identify hepatocytes and their cellular progeny in primary liver cultures. An Albumin-Cre transgenic mouse strain was used that had been characterized extensively and shown to specifically label hepatocytes in both fetal and adult mice (Postic et al., 1999
; Weisend et al., 2009
). Albumin-Cre mice were crossed with ROSA26-mTmG reporter mice which express membranous tdTomato before and membranous EGFP after Cre-mediated recombination (). As expected, the EGFP fluorescence was confined to epithelial cells in freshly isolated liver cultures from these mice (). These cultures were typically composed of ~80% EGFP-positive and ~20% tdTomato-positive cells. However, this ratio declined to 60% EGFP-positive 40% tdTomato-positive cells after one week in culture, implying that hepatocytes were lost and/or other cells outgrew the hepatocytes. Next, we infected these cultures with the three BAM factors and 13 days after dox induction we detected both red and green fluorescent cells with neuronal morphologies (). EGFP-positive cells also expressed the neuronal markers Tuj1 and PSA-NCAM (). Similar results were obtained using an independent reporter allele (ROSA26-Bgeo) (Mao et al., 1999
) where expression of β-galactosidase is induced after Cre-mediated recombination (). These results unequivocally demonstrate that iN cells can be derived from terminally differentiated hepatocytes. We therefore termed these cells hepatocyte (Hep)-iN cells.
iN cells can be derived from terminally differentiated hepatocytes
Hep-iN cells are independent of transgene expression and have acquired functional properties of mature neurons
To determine whether mature Hep-iN cells require sustained transgene expression in order the maintain their phenotype we removed dox from media at different time points after infection. Surprisingly, as few as 5 days of dox treatment sufficed to generate Hep-iN cells, which were present until at least day 22 after addition of dox. Similar results were obtained with MEF-iN cells (). The longer the transgenes were expressed the more iN cells were generated and efficiencies appeared to plateau at around 11 days of dox treatment. While the 3 exogenous factors were strictly dox-dependent the endogenous genes were induced during the reprogramming process (). To investigate whether Hep-iN cells also possessed functional properties of neurons and whether these properties were stable without transgene expression, we performed patch-clamp recordings with cells that were treated with dox for 12 days and cultured for an additional 18 days in dox-free media. Hep-iN cells were identified as EGFP-positive neuronal cells when derived from Albumin-Cre/ROSA26-mTmG mice. We also recorded from Hep-iN cells identified as EGFP/tdTomato-double positive cells when derived from Albumin-Cre/ROSA26-tdTomato/TauEGFP mice (described below, ). The average resting membrane potential of the Hep-iN cells was −50.1±2 mV (n=16). Moreover, spontaneous action potentials were detected in half of the cells (n=8) (). All analyzed Hep-iN cells generated action potentials when depolarized by current injections () and showed fast inactivation sodium current and outward potassium currents (Figure S1G, H
). When Hep-iN cells were FACS-sorted 7 days after dox and cultured together with mouse cortical neuronal cultures for another 4 weeks, postsynaptic responses could be evoked by extracellular stimulation of surrounding neurons (). At holding potentials of −70 mV a small inward current was detected, presumably mediated by AMPA receptors and/or GABAA
receptors. At +60 mV a large outward current was evoked, presumably mediated by NMDA and/or GABAA
Reprogramming efficiencies and kinetics are similar between fibroblasts and hepatocytes
To gain insight into the process of iN cell reprogramming, we first evaluated the cell division frequency after induction of the BAM transgenes in liver cultures by a 5-bromodeoxyuridine (BrdU) incorporation assay. When BrdU was present from the day of infection (i.e. one day before dox) throughout the time of iN cell generation, only 12% of the Tuj1-positive cells at day 13 incorporated BrdU. When BrdU treatment was begun on the day of transgene induction (dox addition) only 1% of the Tuj1-positive cells were BrdU-positive (). Thus, the vast majority of hepatocytes were reprogrammed to iN cells without mitosis.
Efficiencies and timing of Hep-iN cell generation
To address the reprogramming kinetics we generated triple transgenic mice containing the TauEGFP allele together with Albumin-Cre and a ROSA26-tdTomato reporter. In this lineage tracing setting albumin-positive hepatocytes and their progeny constitutively express tdTomato while non-hepatocyte-derived cells remain without fluorescent label ( and S2A
). We established primary hepatocyte cultures from these mice and as expected 13 days after transduction with the BAM factors Tau-EGFP/ tdTomato-double positive Hep-iN cells appeared (). Surprisingly, as early as one day after transgene expression, some infected hepatocytes expressed TauEGFP (). Over time the generation of EGFP-positive cells steadily increased with similar kinetics for hepatocytes and fibroblasts. On day 13, the conversion efficiencies relative to number of plated cells of hepatocytes were similar to postnatal fibroblasts (ca. 6%) but lower than embryonic fibroblasts (ca. 20%) ( and supplementary methods
). When cultured in keratinocyte serum free media (KSFM), a media reported to prevent de-differentiation of cultured hepatocytes (Li et al., 2007
) we observed conversion efficiencies similar to our regular hepatocyte growth media (Figure S1E,F
Finally, we asked whether iN cell reprogramming could be extended to more mature hepatocytes. Following infection with the BAM viruses we could generate iN cells from 1 year old TauEGFP or mice Albumin-Cre/ROSA26-mTomato/mGFP reporter mice (Figure S2C
). Correcting for an assumed infection rate of 30%, we estimated a conversion efficiency of 2.7±1.4% (Figure S2D,E
MEF-iN and Hep-iN cells show global transcriptional remodeling
In order to characterize iN cell formation on the molecular level we determined the gene expression profiles of FACS-purified iN cells 13 and 22 days after dox from hepatocytes, MEFs and TTFs using Illumina’s MouseRef-8 v2.0 Expression BeadChip microarrays (). In addition we profiled the starting populations of Albumin-Cre/Rosa26-tdTomato-positive hepatocytes (FACS-sorted) and MEFs as well as primary neonatal cortical neurons (CN) and neurons derived from fetal (E13.5) forebrain neural progenitor cells (NPCs) 7 and 13 days after differentiation, sorted for TauEGFP expression.
Global transcriptional remodeling during lineage conversion
We first considered only those genes that were differentially expressed (genes with expression changes of at least 3-fold) between hepatocytes, d22 Hep-iN cells, and NPC-derived neurons (13 days after differentiation). Unsupervised clustering identified 3 major clusters of genes and revealed that the vast majority of transcriptional changes in d22 Hep-iN cells approached the levels of primary neurons (Figure S3A
). The largest cluster (cluster B) contained mostly genes with higher expression levels in Hep-iN cells and neurons than hepatocytes. Accordingly, in this cluster the top 8 most significantly enriched gene ontology (GO) terms are associated with neuronal function and development (Figure S3A
). The second largest cluster (cluster C) consisted of mostly those genes downregulated in iN cells and primary neurons (Figure S3A
). Within this cluster many GO terms typical of liver function such as coagulation, wound healing, and inflammatory response were among the most significantly enriched. The analysis also revealed a cluster of genes that were low in hepatocytes and Hep-iN cells but high in neurons (cluster A). This may indicate a group of genes that failed to be induced in iN cells. Indeed, GO terms associated with more mature neuronal function (regulation of membrane and action potential in neuron) were significantly enriched in this cluster (Figure S3A
). However, even more GO terms reflecting glial function were similarly enriched (axon and neuron ensheathment, myelination, lipid biosynthesis, and regulation of action potential). Thus, (i) contaminating glial cells in the NPC-derived cultures may have contributed to many genes in this cluster and (ii) d22 Hep-iN cells may represent a less differentiated state than the primary neurons.
We then performed unsupervised hierarchical clustering of all samples based on 12,275 genes (). Most iN cell samples clustered together with the primary neuron samples indicating that their overall transcriptome is more similar to neurons than to their starting cell types. Surprisingly, NPC-derived neurons were more similar to two iN cell populations (d22 MEF-iN and d22 TTF-iN) than to neonatal cortical neurons. Thus, the transcriptional variability between 2 different primary neuronal populations was greater than between iN cells and a specific population of primary neurons. The various iN cell samples fell into 3 groups: (a) the d22 fibroblast-iN cells, which are most closely approaching primary neurons, (b) d22 Hep-iN cells and d13 fibroblast-iN cells, which are still closer to primary neurons than to fibroblasts or hepatocytes and (c) d13 Hep-iN cells, which are more similar to hepatocytes than primary neurons. This suggests that hepatocytes are harder to reprogram and take longer to induce a complete neuronal program than fibroblasts. The corresponding heatmap showing expression changes of all the 12,275 genes illustrates the genome-wide remodeling of iN cells towards primary neurons (). In addition, Pearson correlation analysis of genes differentially expressed across all samples by at least 4-fold revealed that d13 MEF-iN and Hep-iN cells are much better correlated (R2
=0.1987) than MEFs and hepatocytes (R2
=0.1097) (). Of note, the increased overall correlation is caused by a subset of genes being almost perfectly correlated while the remaining genes appear uncorrelated and higher expressed in Hep-iN cells (). This may suggest that at 13 days the reprogramming factors have induced a portion of the transcriptional program that is similar between Hep- and MEF-iN cells. Our preliminary analysis of expression data directly in response to the BAM factors suggest that the correlated genes are not predominantly direct target genes indicating that the transcription factors induce this pattern through secondary changes (not shown). This is in agreement with the finding that the two different iN cell states become much more similar to each other after 22 days (R2
=0.2824) and the distinct group of uncorrelated genes decreases ( compare d13 with d22 MEF and Hep-iN cell plots). Conversely, d22 Hep-iN cells and hepatocytes appear quite unrelated (R2
=0.02104) while many genes are well correlated between Hep-iN and NPC-derived neurons (R2
=0.2507) ( and S3B
). Thus, based on these data the two iN cell samples are more similar to each other than the donor cell types and iN cells are gradually and increasingly approximating the state of primary neurons.
The BAM factors induce silencing of both MEF and liver-specific transcriptional programs
A key question in lineage reprogramming is whether the induced cell types may represent hybrid phenotypes composed of similarly dominant donor and target cell features or whether reprogrammed cells have efficiently extinguished the donor cell-specific identity. To test the hypothesis whether the BAM pool of transcription factors was capable of silencing the two donor program we first identified a MEF- and liver-specific expression signature by comparing publicly available microarray data from 20 different tissues (Figure S3C
). We then evaluated the expression levels of these genes in iN cells and their donor cells. Strikingly, for both MEF- and Hep-iN cells those donor-specific programs were extensively downregulated. The MEF signature contained 221 probes and 209 (95%) and 201 (91%) were downregulated at least 2-fold in MEF-iN cells at days 13 and 22, respectively. Similarly, the liver-specific signature was composed of 149 probes and 113 (76%) were downregulated at least 2-fold in Hep-iN at day 13 and 126 (85%) at day 22 (). To quantify the extent of silencing we compared expression levels of genes from the liver signature in Hep-iN and neurons. Strikingly, as many as 45% of these liver genes could be considered “turned off” (i.e. showed expression levels lower or up to maximal 2-fold higher than in neurons) (Figure S3D
). Furthermore, we found that Hep-iN cells have completely lost hepatocyte-specific functional properties such as albumin secretion and urea production (Figure S2F-G
We then asked what the extent of reprogramming in Hep-iN cells is on the single cell level. Twenty-eight single Hep-iN cells 32 days after dox treatment from two independently infected cultures, 13 primary Tau-EGFP-positive cortical neurons cultured for 5 days and 13 Albumin-Cre/Rosa26-tdTomato-positive hepatocytes cultured for 6 days were picked and analyzed using Fluidigm dynamic RT–PCR arrays. shows that robust expression of pan-neuronal markers was found in 27/28 Hep-iN cells (β-III-tubulin, Map2, Ncam). Surprisingly, many primary neurons expressed some of the 8 analyzed liver signature genes, illustrating the transcriptional noise of assumed cell type-specific genes. Similar to neurons, Hep-iN cells were randomly positive for one or more liver markers while hepatocytes expressed most of those genes (). When plotting the cells based on how many liver genes were expressed, we found essentially no overlap between hepatocytes and neurons or Hep-iN cells (). On the other hand, the distribution of neurons and Hep-iN cells are overlapping but distinct. Thus, while some Hep-iN cells appear to be indistinguishable from primary neurons there is a trend that Hep-iN cells express slightly more liver genes than neurons. This finding shows that Hep-iN cells do not represent hybrid phenotypes of neurons and donor cell types but possess an epigenetic memory of their cells of origin. The lack of detectable hepatic functional properties suggests that this epigenetic memory has little if any functional consequence.