Metabolic role of dipeptidyl peptidase 4 (DPP4) in primary human (pre)adipocytes

Dipeptidyl peptidase 4 (DPP4) is the target of the gliptins, a recent class of oral antidiabetics. DPP4 (also called CD26) was previously characterized in immune cells but also has important metabolic functions which are not yet fully understood. Thus, we investigated the function of DPP4 in human white preadipocytes and adipocytes. We found that both cell types express DPP4 in high amounts; DPP4 release markedly increased during differentiation. In preadipocytes, lentiviral DPP4 knockdown caused significant changes in gene expression as determined by whole-genome DNA-array analysis. Metabolic genes were increased, e.g. PDK4 18-fold and PPARγC1α (=PGC1α) 6-fold, and proliferation-related genes were decreased (e.g. FGF7 5-fold). These effects, contributing to differentiation, were not inhibited by the PPARγ antagonist T0070907. Vice versa, the PPARγ agonist pioglitazone induced a different set of genes (mainly FABP4). DPP4 knockdown also affected growth factor signaling and, accordingly, retarded preadipocyte proliferation. In particular, basal and insulin-induced ERK activation (but not Akt activation) was markedly diminished (by around 60%). This indicates that DPP4 knockdown contributes to adipocyte maturation by mimicking growth factor withdrawal, an early step in fat cell differentiation. In mature adipocytes, DPP4 becomes liberated so that adipose tissue may constitute a relevant source of circulating DPP4.

accumulation. However, the role of DPP4 in human adipose tissue is currently unclear. Furthermore, recent animal studies with DPP4 inhibitors support the notion that DPP4 may play a functional role within adipose tissue, because DPP4 inhibition has been seen to prevent adipose tissue inflammation and development of glucose intolerance in high fat diet induced obesity in mice 23 . Based on these considerations, we aimed to gain more insight in the role of DPP4 in primary cultured human white preadipocytes and adipocytes.

Results
Expression and release of DPP4 in human white (pre)adipocytes. In order to examine the role of DPP4 in the human adipose tissue, we first measured the expression of this enzyme in primary human white adipocytes at mRNA and protein level during differentiation. Mature adipocytes were obtained by in-vitro differentiation of primary cultured preadipocytes, following the protocol described in the Methods section. DPP4 mRNA was expressed in preadipocytes (i.e. Day 0 of differentiation) and mature adipocytes (up to Day 21 of differentiation) at comparable levels (Fig. 1A). There was a brief dip in DPP4 expression at Days 3 to 9 of differentiation (Day 0, switch to differentiation medium) but there were no major long-term changes.
Comparison of DPP4 expression level in preadipocytes and adipocytes with various other human tissues (Fig. 1B) revealed that preadipocytes and mature adipocytes express 4 th and 6 th highest level of DPP4, respectively (highlighted in Fig. 1B).
DPP4 protein weakly increased with differentiation as revealed by Western blotting and consecutive densitometric quantification (Fig. 1C), but statistical significance was not reached.
In order to test whether (pre)adipocytes could be a source of soluble DPP4, we analyzed cell culture supernatants, collected at different time points of differentiation, by ELISA (Fig. 1D). Preadipocytes liberated DPP4 to a low extent only (around 1 ng/ml), but DPP4 release markedly increased during differentiation. Mature adipocytes released up to 11-times more DPP4 than non-differentiated cells. A linear regression analysis revealed statistical significance. It should be noted that DPP4 is not stored in vesicles before release (instead, the extracellular part of this membrane protein is cleaved off) so that cellular DPP4 content does not need to increase when liberation of DPP4 increases.
In a next step, we questioned which events could influence DPP4 release. Cellular triglyceride content was a candidate because triglyceride storage and liberation is a prominent function of adipocytes. In order to change the triglyceride content of the cultured adipocytes, we induced lipolysis by treatment with the cAMP-mimetics forskolin (10 μM) or DBcAMP (100 μM) for 24 h (Fig. 1E). Triglyceride degradation was verified by measuring the increase of glycerol content in the cell culture supernatants. DPP4 release was measured by ELISA 24 h after induction of lipolysis. Leptin secretion was also determined for comparison. It turned out that leptin secretion was significantly reduced following lipolysis whereas no influence on DPP4 release was detected (Fig. 1E).
For further characterization, we determined the intracellular localization of DPP4 protein in preadipocytes. Western blotting following cell fractionation revealed a strong DPP4 signal in the membrane fraction (Fig. 1F). The cytosolic fraction showed several weak bands, probably due to unspecific binding of the antibody. No signal could be detected in the nucleus and in the cytoskeleton. In accordance with this observation, detection of DPP4 protein in preadipocytes by immunofluorescence with z-stack analysis (Fig. 1G) revealed DPP4 (green signal) localization primarily in the outer cell regions, i.e. in a typical appearance of membrane proteins.
Taken together, DPP4 is expressed in preadipocytes and adipocytes, is located primarily in the cell membrane from which it becomes increasingly released during maturation.
Gene expression profile after DPP4 knockdown. As described above, DPP4 was highly expressed in preadipocytes but was hardly released from these cells. This implies a different function of DPP4 in preadipocytes as compared to mature adipocytes. For closer investigation, we produced preadipocytes with a stable knockdown of DPP4 expression, achieved by lentiviral shRNA (see Methods section). Successful lentiviral transduction was verified with a GFP construct (Supplementary Figure S1A). Unspecific effects of the lentiviral vector were excluded by using a non-targeting shRNA ("sh-control" or "SHc") as negative control in each experiment. Furthermore, two other independent shRNA constructs, also directed against DPP4, were tested and found to induce the same set of genes (Supplementary Figure S1B). Knockdown of DPP4 was verified at mRNA ( Fig. 2A) and protein level (Fig. 2B).
No clear hints were available what the function of DPP4 in preadipocytes could be so that we started our study with a screen for changes in gene expression using a whole genome oligo microarray (Agilent). A heat map of four replicate experiments visualizing genes that were altered at least 5-fold is shown in Fig. 2C. Salient genes, altered at least two-fold and forming functional clusters are listed in Table 1. The most pronounced changes were found in genes involved in lipid metabolism (with an up-regulation in most cases) and in some proliferation related genes (down-regulated in most cases) as summarized in Table 1. For example, a 5-to 24-fold up-regulation of FABP4, PDK4 and PPARγ C1α was observed.
The transcription factor C/EBPε , a relative of C/EBPα and C/EBPβ , was induced 20-fold. C/EBPα and C/ EBPβ are known to be involved in adipocyte differentiation. C/EBPε was initially detected in lymphoid and myeloid cells based on its structural similarity to C/EBPα and C/EBPβ 24 , and is assumed to regulate their differentiation 25 . The function of C/EBPε in preadipocytes is unknown but, due to its structural similarity, C/EBPε could mimic the effects of C/EBPα and/or C/EBPβ . Other transcription factors were also increased, such as members of the KLF family (e.g. KLF15, − 5, − 2). Among the genes induced or suppressed at least twofold, four functional clusters became obvious (Table 1), namely lipid metabolism, proliferation, structural genes including cell-cell contact and cell migration.
Representative metabolic and proliferative genes were selected for further investigation by quantitative PCR. We studied the time course of expression and the interaction with other agents known to be involved in adipocyte differentiation. The time course of PPARγ C1α and PDK4 expression after knockdown is shown in Fig. 2D Figure S2C. PCR data are presented as mean ∆ CP values (normalized to GAPDH, relative to SHc) ± SEM (left y-axis) and calculated fold change values vs. control (right y-axis), n ≥ 5. Statistical analysis was done by one-way ANOVA with Dunnett post-test; **p < 0.01 vs. negative control. Part (C) At least 5-fold changes in gene expression resulting from DPP4 knockdown, measured by whole genome DNA array hybridization, are visualized in a heat plot. Hybridization was performed in two-color mode; each line represents the difference between a DPP4 knockdown and sh-control sample. The four lines represent four biological replicates. Up-regulated genes in DPP4 knockdown compared to control are marked in green, downregulated genes in red. The color intensity indicates the expression level of the respective gene. Part (D) Changes in the expression of two representative genes (PPARγ C1α and PDK4) over time after infection were followed by quantitative PCR. Data represent mean ∆ CP values (vs. GAPDH) ± SEM (left y-axis) and calculated fold change values vs. control (SHc at Day 0) on the right y-axis, n ≥ 3. Statistical analysis was done by t test; *p < 0.05; **p < 0.01 vs. control. Part (E) shows a Western blot confirming the up-regulation of PPARγ 1Cα on protein level when DPP4 is suppressed. The blot was cropped for clarity. The entire lanes are shown in Supplementary Figure S2D. The results for DPP4, PPARγ 1Cα and the loading control α -actinin from cells infected with shcontrol vector ("SHc") and sh-DPP4 vector ("DPP4 KD"), respectively, are shown as indicated in the figure. The gene PPARγ C1α , also known as PGC1α , is a transcription regulator and appears to have a crucial role in cellular energy metabolism, in particular in mitochondrial biogenesis 26 . Therefore we confirmed its up-regulation also on the protein level by Western blotting (Fig. 2E).
DPP4 is a multifunctional protein which actions beyond peptidase activity. Thus, we tested whether the observed effects by the knockdown could be due to the peptidase activity. The latter can be inhibited by sitagliptin (10 μM). No effect of sitagliptin on the expression of the genes responsive to DPP4 knockdown was observed (not shown). This indicates that a non-peptidase function of DPP4 is responsible for the observed regulation of gene expression.
Comparison of DPP4 knockdown with PPARγ effects. The alterations in gene expression induced by DPP4 knockdown imply that the latter contributes to the differentiation of the preadipocytes as outlined above. A well-established inducer of adipocyte differentiation is the transcription factor PPARγ 27 . Therefore, we asked whether the effects of DPP4 knockdown described above could be mediated by PPARγ . We compared the effects of DPP4 knockdown with the effects of treatment with the PPARγ agonist pioglitazone (100 μM for 72 h, Fig. 3A). The sets of genes that were induced by DPP4 knockdown and pioglitazone treatment, respectively, were similar since both sets were indicative for adipocyte differentiation. However, there were marked differences. A main feature of the pioglitazone effect was the strong induction of FABP4. With DPP4 knockdown, FABP4 induction was much weaker. In contrast, DPP4 knockdown strongly and consistently induced PPARγ C1α and APOE which were hardly affected by pioglitazone (Fig. 3A). Furthermore, expression of the growth factor FGF7 was suppressed by DPP4 knockdown but was not affected by pioglitazone. For further confirmation of independent actions of DPP4 and PPARγ , we employed the PPARγ antagonist T0070907 (10 μM). The changes in gene expression were similar with DPP4 knockdown alone and with DPP4 knockdown plus T0070907 (Fig. 3A). Thus, T0070907 did not alter the gene expression induced by DPP4 knockdown.
A further difference between the action of DPP4 knockdown and PPARγ activation was revealed by determining lipid accumulation. Lipid droplets (marked by arrows in the figure) were observed after pioglitazone stimulation but not after DPP4 knockdown (Fig. 3B).
T0070907 alone had no major effects on the DPP4 knockdown expression profile except for some up-regulation of PDK4 (Fig. 3C, right group of bars). In order to confirm that T0070907 had no unspecific effects (i.e. actions beyond PPARγ inhibition) we compared the effect of T0070907 on gene expression with the effect of PPARγ knockdown (Fig. 3C, mid group of bars). Overall, the pattern was similar except for an up-regulation of PLIN1 by PPARγ knockdown but not by T0070907 treatment.

Effect of DPP4 knockdown in later stages of differentiation. DPP4 knockdown elicited changes
in gene expression in preadipocytes. Thus, we investigated the effect of DPP4 knockdown also in later stages of adipocyte maturation. Preadipocytes stably transduced with DPP4 shRNA were differentiated according to the standard protocol (see Methods section), and the expression of the genes of interest was studied by RT-PCR at Days 0, 6 and 12 of differentiation (Fig. 4). For comparison, the differentiation protocol was also performed with cells expressing non-target control shRNA.
Compared to Day 0 of differentiation, the effect of DPP4 knockdown diminished during differentiation despite DPP4 mRNA remained suppressed. At Day 0 of differentiation (i.e. before switching the cells to differentiation medium), the expression rates of the genes shown were higher (or lower in case of FGF7) in DPP4 knockdown cells compared to sh-control cells. The differentiation process caused an increase of the metabolic genes (FABP4, PDK4, PPARγ C1α , PLIN1 and APOE). This effect was more pronounced in the sh-control cells so that after differentiation (Day 12) the expression levels of these metabolic genes were virtually identical in the DPP4 knockdown and in the sh-control cells (Fig. 4). The growth factor FGF7 was decreased in preadipocytes in response DPP4 knockdown, but also for this gene the expression levels became similar in DPP4 knockdown and sh-control cells after differentiation (Fig. 4).

Effect of DPP4 knockdown on intracellular signaling. For closer investigation of the mechanisms
by which DPP4 knockdown exerts the described effects on gene expression, we studied the activation of protein kinase signaling pathways (Fig. 5). Growth factor withdrawal is the first step of adipocyte differentiation in vitro, and it was reported that in vivo an autocrine EGF-related growth factor, Pref-1, prevents differentiation via activation of ERK 28,29 . In cultured preadipocytes we observed a basal activity of the ERK pathway, measured as phosphorylated ERK (pERK) by Western blotting. Preadipocytes express insulin receptors (Fig. 5A); insulin receptor expression was not influenced by DPP4 knockdown. ERK phosphorylation was markedly enhanced by stimulation with insulin (100 nM for 10 min). After knockdown of DPP4, insulin-induced ERK phosphorylation was significantly weaker (Fig. 5A; densitometric quantification in Fig. 5B). In contrast, activation of the pAkt Table 1. Effects of DPP4 gene expression knockdown in human primary preadipocytes. Whole genome oligo microarray (Agilent) analysis was used to investigate changes in the expression profile of human primary preadipocytes transduced with shRNA against DPP4. Functional clusters of the most strongly altered genes are displayed, with the data represent the log (ratio) and mean fold change ± SEM of the gene expression after knockdown of DPP4 vs. control. Mean data result of 4 replicate measurements.
In line with the described effects on growth factor signaling, DPP4 knockdown prevented further proliferation of the preadipocytes as measured by cell counting over time (Fig. 5C).
Taken together, DPP4 knockdown in preadipocytes diminished the ability of insulin and probably other growth factors to activate the ERK signaling pathway. This mimics growth factor withdrawal, leads to growth arrest and could thereby contribute to initiate the first step of differentiation.

Discussion
The various functions of DPP4 have been widely discussed, among others in the fields of immunology, (neuro-) endocrinology and glucose homeostasis 8,10 . However, the role of DPP4 in human adipose tissue is still unclear.
Our results now revealed a strong expression of this gene in human white preadipocytes and adipocytes and revealed a possible contribution of DPP4 to the adipocyte differentiation process. Furthermore, mature adipocytes were identified as a potential source of circulating DPP4.
Adipocyte maturation is a complex process and involves several different mediators and signaling pathways [30][31][32][33] . Among these are the two master regulators PPARγ and the C/EBP family 32,34 . Our knockdown experiments revealed changes in the expression of functional gene clusters indicative for adipocyte differentiation.
Investigation of signaling pathways identified a potential mechanism by which DPP4 knockdown could contribute to differentiation. It became obvious that basal and insulin-induced ERK phosphorylation was attenuated. In contrast, activation of the Akt pathway by insulin was not affected, arguing for a selective action of DPP4 on growth factor signaling via ERK.
It should be noted that insulin probably has a dual role in respect to adipocyte differentiation 32 . On one hand, insulin promotes differentiation and is a component of the differentiation medium. For this effect activation of the pAkt signaling pathway appears to be relevant. On the other hand, by activation of ERK insulin behaves like a growth factor and may thereby counteract the onset of differentiation. The role of ERK in adipocyte differentiation is not fully clear 32 , but in the case of the EGF-related growth factor Pref-1, which acts on preadipocytes in an autocrine way, it was clearly shown that ERK activation by Pref-1 prevents differentiation 28,29 .
The effects of DPP4 knock-down were not influenced by inhibition of PPARγ , an important player in adipocyte maturation but acting at a later stage of this process. Accordingly, the set of genes induced by DPP4 knockdown differed from the set induced by PPARγ . The action of DPP4 at an early step in the differentiation process also explains why DPP4 knockdown, in contrast to PPARγ activation, did not promote triglyceride accumulation because the latter most likely is a late event in adipocyte maturation.
Beside of metabolic genes, genes encoding extracellular matrix proteins and proteins being involved in cell-cell interaction and migration were altered by DPP4 knockdown. A link between the extracellular matrix composition and the differentiation competence involving the action of matrix metalloproteinases (MMPs) was described already 33 . Moreover, a role for DPP4 in the modulation of the extracellular matrix could be shown in different studies 35 .
In accordance with our findings, a recent publication by Han et al. 36 identified other members of the DPP4 family, DPP8 and DPP9, as players in adipocyte differentiation. This group used a permanent mouse cell line (3T3-L1), and in contrast to the effects of DPP4 observed in our work, blocking or knockdown of DPP8 and DPP9 inhibited differentiation with involvement of PPARγ .
One strongly induced gene in response to DPP4 knockdown was PPAΡ γ C1α . This gene, also known as PGC1α , is related to mitochondrial function and it is necessary for brown fat development 26,37 . It acts as a master regulator of mitochondrial biogenesis in mammals and therefore participates in energy balance. However, other genes characteristic for brown or "beige" fat cells, such as UCP1 37,38 , were not increased.
In line with previous studies, we detected increased DPP4 protein expression and release in a maturation-dependent way 18,39 . This was also observed during monocyte differentiation to dendritic cells 14 . However, in our study adipocyte maturation predominantly affected DPP4 liberation; DPP4 expression changed only to a minor extent. This is in contrast to findings by Das et al. 39 in the mouse cell line 3T3-L1 where a strong increase in DPP4 protein expression during adipocyte differentiation was observed. We could hardly measure a DPP4 release in the supernatants from preadipocytes but observed a marked increase during adipogenesis, pointing to DPP4 release as an important function of mature adipocytes. This is in line with the recently recognized role of the adipose tissue as an endocrine organ 21 . The mechanism underlying increasing DPP4 release during maturation is not clear, and as it is not yet established which factors trigger DPP4 release. Another prominent adipokine, leptin, becomes released in dependence of the adipocyte triglyceride content. No such dependency was observed for DPP4. Taken together, DPP4 was found to be highly expressed in adipose cells, and its knockdown contributes to differentiation of human preadipocytes, obviously at an early stage and in a PPARγ -independent way. In preadipocytes DPP4 appears to play a different role than in mature adipocytes where it becomes released and thereby can influence glucose metabolism via incretin (e.g. GLP1) cleavage.

RNA Extraction and Real-Time PCR.
Cells were harvested with lysis buffer from the RNeasy Total RNA Extraction Kit (Qiagen, Hilden, Germany), and RNA was isolated according to the manufacturer's protocol. Complementary DNA was synthesized using Reverse Transcriptase Kit (Roche, Mannheim, Germany). Real-Time PCR reactions were performed with the LightCycler SYBR Green Master mix (Roche, Mannheim, Germany) on a Roche LightCycler 480 instrument with denaturation for 10 s at 95 °C and annealing/extension for 105 s at 72 °C. The nucleotide sequences of the PCR primers used are given in Table 2. The expression levels of the target mRNAs were determined by the crossing point (CP) method. The results were corrected for primer efficiency and normalized to the housekeeping gene glyceraldehyde-3-phosphate dehydrogenase (GAPDH). DNA Array Hybridization. Cells were harvested with lysis buffer as described above to extract total RNA.
Quality of RNA was controlled by the Agilent 2100 Bioanalyzer platform (Agilent Technologies, Santa Clara, CA, USA); RNA Integrity Number (RIN) was between 9.8 and 10. Amplification and labelling (Cy3 and Cy5) were performed using the Low RNA Input Linear Amp Kit (Agilent Technologies, Santa Clara, CA, USA). Hybridization was conducted on Agilent Whole Genome Oligo (60-mer) 4 × 44K microarrays with the Agilent Gene Expression Hybridization Kit. Agilent's feature extraction Software was used to determine spot intensities and Cy5/Cy3 ratios after background subtraction with ratios displaying the expression level in KD compared to control in a logarithmic scale. Mean fold changes were calculated from four replicate measurements. The heat map was created by Gene Spring Software (Agilent Technologies, Santa Clara, CA, USA).