System for single-cell analysis of human islets
To study the heterogeneity of human islets at the single-cell level, we performed single-cell gene expression analyses of six different batches of human islets via the Fluidigm microfluidic platform (Fig. 1a and Supplementary Fig. 1A). Human islets and the reciprocal non-islet fractions were isolated in Canada and shipped to Singapore as soon as possible. On average, they were analyzed after being ex vivo in CMRL or MIAMI media for 3–6 days, nearly immediately upon arrival to reduce prolonged impact of ex vivo culture on islet transcriptome. We began our gross quality control by analyzing bulk transcript expression from the islet and non-islet fractions of five batches of human islets. qPCR analyses for mature pancreatic endocrine cell markers INS, GCG, SST, GHR, PPY, GCK, PCKS1, PCKS2, CHGA, CHGB, SYP, and KCNJ11 clearly demonstrated higher transcript expression in islets as compared to the non-islet fractions (Supplementary Fig. 1B). FACS analyses on human islets confirmed the presence of ~59.8% of INS+ cells (INS antibody tested using MIN6 mouse β-cell line (Supplementary Fig. 1C), a low percentage (12.3%) of GCG+ cells, 8.5 of SST+ cells and a relatively high percentage (24.4%) of PPY+ cells (Supplementary Fig. 1D). Subsequently, ~300 human islets were picked, trypsinized into single cells and stained for live cell (using calcein-AM) prior to cell capture in the microfluidic chip (Fig. 1a). For each chip, 48 randomly chosen live cells were analyzed by nested quantitative reverse transcription PCR (RT-qPCR) analyses (Fig. 1a).
Polyhormonal signatures in single human islet cells
The five pancreatic endocrine cell types, β, α, δ, ε, and PP cells express and secrete INS, GCG, SST, GHR, and PPY hormones, respectively. Concordant with earlier FACS analyses (Supplementary Fig. S1D), we detected ~63% INS+ cells and ~9% GCG+ cells in our single-cell gene expression analyses (Fig. 1b). As expected, a low percentage (~5%) of GHR+ cells were detected. Surprisingly, ~40% SST+ and ~27% PPY+ cells were detected, consistent across six different batches of human islets where we analyzed a total of 281 single cells (Fig. 1b). Congruent with our data, Wang et al.12 also observed high-expression levels of SST and PPY transcripts although they were expected to be in very low abundance12. This reproducibility provided confidence that the phenomenon was real at the single islet cell level.
Next, we evaluated the proportion of islet cells that were either not expressing endocrine hormonal transcripts or, containing one or more transcripts per islet cell. Twenty percent of the cells were found not to express any of the five hormonal transcripts and only 34% were found to be monohormonal positive. Surprisingly, a total of 46% of the islet cells were found to express two or more hormonal transcripts per cell (Fig. 1c). These data were strikingly consistent with Katsuta et al.13 that reported 45% of rodent β-cells to express multihormonal transcripts13. In addition, Chiang and Melton14 have also reported the detection of multiple endocrine-expressing cells such as Gcg+Ppy+ cells in adult rodent islets14. Indeed, Wang et al.12 also recently reported the presence of endocrine cells, albeit rare, to have conflicted expression profiles12.
To better understand this human islet cell heterogeneity, we parsed the single-cell data into single-hormone, double-hormone, and triple-hormone transcript-positive cells (Fig. 1d–f). From the heatmap, it can be appreciated that there is a small proportion of INS+GCG+ cells (due to the small percentage of GCG+ cells) but a far greater proportion of INS+SST+, INS+PPY+, and INS+SST+PPY+ cells (Fig. 1e, f). Similarly, Katsuta et al.13 reported a small proportion (11%) of Ins+Gcg+ cells but a greater proportion (29%) of Ins+Sst+ cells and (29%) Ins+Ppy+ cells among the single rodent β-cells13. To verify these polyhormonal signatures at the protein level, we performed co-immunostaining for C-pep (surrogate for secreted INS), GCG and SST in human islet sections. Gratifyingly, we detected several C-pep+GCG+ and C-pep+SST+ cells (Fig. 1g), confirming our single-cell gene expression analyses. These antibodies have also been validated in human pancreas sections (Supplementary Fig. S1E). The co-expression of INS and GCG in a single-cell is not all that surprising since ~1% of INS+GCG+ cells have been observed in the adult human pancreas15.
As human fetal pancreata have been reported to harbor multihormone-containing cells16,17, speculatively, it may be possible that these multihormonal transcripts either (1) represent historical signatures when the islet cells were immature or (2) a sign of de-differentiation into other hormonal cell types. Since these are fully differentiated and mature human islet cells, these polyhormonal cells could possibly suggest an inter-conversion and transition from one cell type to the other during 3–6 days of ex vivo culture. After β-cell ablation in rodents, α-18 or δ-cells19 can actually convert to β-cells. Conversely, β-cell de-differentiation into multipotent precursor cell states20 or into a new non-physiological state21 is also possible (at least in rodents). This speculative β-cell de-differentiation may imply altered gene expression and/or function, eventually leading to decreased functional β-cell mass and/or impaired glucose-stimulated insulin secretion (GSIS).
Loss of β-cell features
We then sought to thoroughly review the β-cell profile for its functional machinery signature. Although there was excellent overlap for INS+ (~63%) and KCNJ11+ (~74%) cells (the KCNJ11 gene encodes subunits of the KATP channel in β-cells to facilitate the exocytosis of INS under hyperglycemic conditions) (Fig. 2a), there was a very low-percentage of GCK+ (glucose sensor; two different pairs of qPCR primers used), PCSK1+ and PCSK2+ (convert proinsulin to INS intermediates) cells (Fig. 2a). Some CHGA+, CHGB+ (secretory granule genes), SYP+ (synaptic vesicle gene), and GLUT1+ (major glucose transporter in human β-cell) cells were detected but they only comprised a fraction of the ~63% of INS+ cells (Fig. 2a). The heatmap for these transcripts mapped upon INS+ and INS− cells (Supplementary Fig. S2A) also reflected a general loss of gene expression involved in β-cell functional machinery.
We then evaluated the expression profile of a few cardinal β-cell transcription factors. In general, HLXB922, PDX123, NKX6.1, and MAFA transcripts were expressed at a higher level when comparing human islet versus non-islet fractions (Fig. 2b). However, they only comprised a small percentage of the INS+ cells (Fig. 2c and Supplementary Fig. S2B). Despite using two different pairs of qPCR primers for PDX1, its percentage remained consistently low (Fig. 2c). The loss of Nkx6.1 results in decreased Pdx1 and MafA expression, and an acquisition of a δ-cell-like fate24,25. This could partly explain the high proportion of INS+SST+ cells in these analyses (Fig. 1e, f). MafA is required for mature β-cell function and the expression of Pdx1, Nkx6.1, and Pcsk126,27. The loss of MafA is also known to result in the de-differentiation of β-cells28. Hence, the low frequency of MAFA+ cells could also contribute to the “gradual loss of β cell identity”. We note that there are some INS− cells that continue to express some β-cell-specific transcription factors (Supplementary Fig. S2B). They could be “empty β cells” that no longer contained INS but still retained some β-cell transcriptional footprint.
Following our transcriptional analyses, co-immunostaining for C-pep and CHGA confirmed the presence of C-pep+ β cells that no longer co-express CHGA secretory granule protein (Fig. 2d). Some C-pep+ cells were also found to be negative for β-cell transcription factor PDX1 (Fig. 2d; antibodies validated in human pancreas sections (Supplementary Fig. S1E)). Collectively, our single-cell transcript and protein analyses indicate the loss of β-cell features in ex vivo human islet cells.
Analyses of select α- and δ-cell markers
Next, we briefly evaluated several known α- and δ-cell markers. α-cell transcription factors IRX2 and ARX8,29 are expressed at higher levels in islet as compared to non-islet fractions (Fig. 3a), suggestive of α-cell-specificity. Brn4 may play some roles in α-cell specification30,31 but its lack of human islet fraction-specificity suggests that its role is not exclusive to α-cells (Fig. 3a–c). While β- to α-cell de-differentiation is a prevailing ideology in the pancreatic islet cell field32, the very low proportion of GCG+, IRX2+, or ARX+ cells (Fig. 3b, c) do not seem to support the case for any extensive β- to α-cell de-differentiation, at least in normal ex vivo human islets. δ-cell marker CCKBR33,34 appeared to be more specific to the islet fraction as compared to HHEX35 transcript expression (Fig. 3d). Nonetheless, the proportion of CCKBR+ or HHEX+ cells is still lower than SST+ δ-cells (Fig. 3e, f).
Pancreatic progenitor profile in islet cells
Many pancreatic transcription factors are known to play multiple roles during pancreatic development. However, their relative abundance (measured by qPCR analyses) in mature single human islet cells is yet unclear. Our profiling revealed a high proportion of cells expressing HNF4A, HNF1A, and GATA636 transcripts as compared to FOXO1, KLF11, PAX4, PAX6, GATA4, or RFX6 pancreatic transcription factors (Supplementary Fig. S3A). The higher percentage of HNF1A+ as compared to HNF4A+ cells could suggest a more predominant role for HNF1A in mature human islet cells. HNF4A and HNF1A transcripts were also confirmed to be expressed at higher levels in islets as compared to non-islet fractions37 (Supplementary Fig. S3B).
Next, we evaluated pancreatic progenitor (HNF1B and SOX9) and endocrine progenitor (NGN3 and NEUROD1) markers in the islet and non-islet fractions (Supplementary Fig. S3B). HNF1B and SOX9 are known to be expressed in pancreatic ducts38,39. Hence their transcripts can be detected at higher levels in the non-islet fractions (Supplementary Fig. S3B). In contrast, NGN3 and NEUROD1 are expressed at higher levels in the islet fractions, in accordance with their pro-endocrine roles in islet differentiation39 (Supplementary Fig. S3B). Importantly and surprisingly, we found a high percentage of HNF1B+, SOX9+, NGN3+, and NEUROD1+ cells in our single-cell gene expression analyses (Fig. 4a). This is unprecedented given that these progenitor transcription factors are not known to be highly expressed in mature human islet cells.
Correspondingly, this led us to hypothesize that the ex vivo culture of human islet cells could have resulted in some form of “de-differentiation” and a “re-expression” of progenitor markers such as HNF1B, SOX9, NGN3, and NEUROD1. This phenomenon was recently observed in cultured human islets40. To determine whether there was a correlation between the expression of pancreatic progenitor transcription factor signatures and that of the hormonal transcripts, we compiled the data and expressed them in a heatmap. Many INS+, SST+, and PPY+ cells were found to co-express five to six of the pancreatic progenitor transcription factors PDX1, PTF1A, HNF1B, SOX9, NGN3, and/or NEUROD1 (Fig. 4b). This was also true for the co-expression of three to four of these transcription factors (Supplementary Fig. S3C). Co-immunostaining for C-pep+ β cells and SOX9 or NEUROD1 then revealed some rare double-positive cells (Fig. 4c; antibodies validated in human pancreas sections (Supplementary Fig. S1E)). Collectively, these data are strongly suggestive of a state of transition of the human islet cells, re-expressing pancreatic progenitor signatures and possibly contributing to the presence of multihormonal transcripts. The large percentage of SST+ cells could suggest a transition between δ- and β-cells as reported in Chera et al.19.
Overlap in exocrine and progenitor signatures
Subsequent characterization of pancreatic exocrine profile first confirmed that PTF1A41, MIST142, NR5A243,44, KRT19, AMY (amylase), and PRSS1 (trypsin) transcripts were expressed at higher levels in the non-islet as compared to the islet fraction (Supplementary Fig. S4A), in agreement with the known functions of these pancreatic exocrine genes. Our single-cell gene expression analyses surprisingly revealed a high percentage of PTF1A+, MIST1+, NR5A2+, KRT19+, AMY+, and PRSS1+ cells (Fig. 4d), consistent across six different batches of human islets. Similarly, we questioned whether there was a correlation between the expression of pancreatic exocrine genes and pancreatic progenitor transcription factor signatures. Heatmap analyses clearly indicated that there were many single human islet cells co-expressing five to six pancreatic exocrine genes with pancreatic progenitor transcription factors HNF1B, SOX9, NGN3, and NEUROD1 (Supplementary Fig. S4B). Co-immunostaining for NEUROD1+PRSS1+ and NEUROD1+AMY+ cells interestingly revealed some double-positive cells in the non-islet fractions (Supplementary Fig. S4D; antibodies validated in human pancreas sections (Supplementary Fig. S1E)). Altogether, these data suggest a possible progression from a transient progenitor state toward pancreatic exocrine cell fate or vice versa.
Overlap in endocrine and exocrine signatures
The presence of pancreatic progenitor and exocrine gene profile raised the possibility of human islet cells co-expressing transcripts of both endocrine and exocrine “mixed” identity as compared to “contaminating” exocrine cells in the islet fraction. To evaluate this hypothesis, we clustered single human islet cells expressing hormonal transcripts with those that expressed pancreatic exocrine genes and presented the data in a heatmap (Fig. 4e and Supplementary Fig. S4C). Specifically, we found ~21% (48/233) of cells to co-express seven to ten pancreatic endocrine and exocrine genes, and ~49% (115/233) of cells to co-express four to six endocrine and exocrine genes. Subsequent co-immunostaining analyses revealed some rare INS+AMY+ cells in the human islet sections (Fig. 4f). However, we struggled to find INS+ cells in the non-islet fractions (data not shown). Our analyses on at least two different batches of islet/non-islet fractions did not reveal cells double-positive for INS and PRSS1 (Supplementary Fig. S4D).
While this is conceptually troubling, INS+AMY+ cells have actually been reported in adult human pancreases45. Wang et al.12 also acknowledged the presence of cells of a “conflicted endocrine/exocrine nature” but only attributed this phenomenon to islet samples from children and did not investigate further12. Based on our data and the above-mentioned studies, we believe that this phenomenon is indeed genuine in adult human islet samples. These cells could be acinar β-cells, pancreatic exocrine cells transiting into INS+ cells or INS+ cells converting into a mixed exocrine state. Nonetheless, among the INS+ cells, there are ~22% (52/233) of INS+AMY− cells that are presumptively the “normal and good” human β-cells remaining after 3–6 days of ex vivo culture. In summary, we report the presence of de-differentiation signatures (rare polyhormonal cells, decreased β-cell and increased progenitor identity) in ex vivo human islets (Fig. 5).