The Diabetic Phenotype in HNF4A Mutation Carriers Is Moderated By the Expression of HNF4A Isoforms From the P1 Promoter During Fetal Development

RNA samples were obtained from kidney (pooled from six Caucasian women, 28–52 years old) (Clontech, Oxford, U.K.), liver (51-year-old Caucasian man), total pancreas (pooled from five Caucasian men and women, 24–77 years), fetal pancreas (pooled at 19/27 weeks’ gestation), colon (pooled from five Caucasian men and women, 20–55 years), small intestine (pooled from five Caucasian men and women, 20–61 years), stomach (50-year-old Caucasian man), and isolated human islets (two independent samples) (National Disease Resource Interchange [NDRI], Philadelphia, PA). The collection of early human fetal material and its ethical approval have been described previously (36,37). Material from eight human fetuses assessed for gestational age according to foot length was studied. Mouse islet RNA was a gift of Dr. Anna Gloyn. All RNA samples were DNAse treated before reverse transcription using the TURBO DNAse kit (Ambion, Huntingdon, U.K.).

Reference sequences corresponding to the six known 5′and 3′ HNF4A variations were identified from the NCBI Database (http://www.ncbi.nlm.nih.gov). The ubiquitously expressed β-2-microglobulin (β2M) gene was selected as a control on the basis of its constant expression levels across the tissues and developmental stages studied. An assay for this gene (assay Hs00187842) was purchased by Assays-on-Demand from Applied Biosystems (Foster City, CA). Custom real-time PCR assays were obtained from the Assays-by-Design service available from Applied Biosystems. Probe and primer sequences are given in supplementary Table 1 (available in an online appendix at http://dx.doi.org/10.2337/db07-1742). Where possible, probes were chosen to span introns to ensure amplification of cDNA rather than genomic DNA. Assays were validated by standard curve analysis of serial 1:2 dilutions of pooled liver, kidney, and pancreas cDNA. The efficiency of each assay was assessed by reference to the slope of the corresponding standard curve. The correlation between cDNA concentration and crossing point (r² value) was assessed from the fit of the crossing point corresponding to each dilution point to the curve.

Complementary DNA (cDNA) was synthesized from 3 μg of the total RNA using the Thermoscript RT-PCR system (Life Technologies, Paisley, U.K.) using 50°C as the incubation temperature. Real-time PCR reactions to determine β2M and HNF4A expression levels were carried out using the ABI Prism 7000 platform (Applied Biosystems). Each sample was amplified in triplicate to ensure accuracy of quantification. Where multiple samples per tissue were tested, each sample was from a separate mRNA extraction and reverse transcription. PCRs contained 10 μl TaqMan Universal Mastermix (no AMPerase) (Applied Biosystems), 0.9 μmol/l each primer, 0.25 μmol/l probe, and 2 μl cDNA reverse transcribed as above in a total volume of 20 μl. PCR conditions were a single cycle of 95°C for 10 min followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. Human fetal pancreas RNA was subject to a pre-amplification step before quantification of targets using the Applied Biosystems TaqMan PreAmp Master Mix kit (Applied Biosystems), which allows the study of multiple targets from small amounts of tissue.

Expression levels of β2M and HNF4A isoforms were calculated by reference to the average crossing point of the triplicate samples. For each RNA sample, the difference between the average crossing point (Ct) obtained from β2M amplification (Ct^β2M) and the average crossing point obtained from the test probe set (Ct^test) was termed the ΔCt value. ΔCt values were calculated for each isoform. The level of each transcript relative to β2M levels could then be calculated from the equation 2^−ΔΔCt (38), where ΔΔCt is the ΔCt value of the test transcript (ΔCt^test) normalized to a reference transcript, which was taken to be the levels of HNF4A for the probe set in question found in human liver (ΔCt^ref). Transcripts were identified as originating from the P1 promoter or the P2 promoter using probe set 1, which identified a) HNF4A1–4A3 (P1), b) HNF4A4–4A6 (P1), and c) HNF4A7–4A9 (P2) transcripts. The occurrence of each 3′ variant was then determined using probe set 2, which identified a) HNF4A1/4A4/4A7, b) HNF4A2/4A5/4A8, and c) HNF4A3/4A6/4A9 transcripts. Where expression was exclusively from one promoter, the identity of each isoform could be determined in this manner by combining data.

Data were gathered from families that had been referred for testing in our laboratories in the period between 1996 and 2006 (n = 84) and from families reported in the literature for which age of diagnosis data were available (n = 106) (8). We studied the possible effect of alternate mRNA processing on phenotype in a total of 45 different mutations found in 190 subjects from 58 families, shown in online supplementary Fig. 1. Of these, 50 (26%) patients had mutations in the P2 promoter or exon 1D, 125 (66%) grouped in exons 2–8, and 15 (8%) had mutations in exons 9 and 10.

We correlated the age at diagnosis with mutation position relative to the known structure of the HNF4A isoforms. To confirm that any effect noted was due to the effect of mutation on different HNF4A isomers and was not related to differences in mutational mechanism, we also correlated age at diagnosis with mutation type (missense or premature termination codon forming) and with the functional domain structure of the HNF-4α protein by Mann-Whitney U analysis. To account for relatedness between family members, we also carried out additional analysis using generalized estimating equation (GEE) analysis. Age of diagnosis was used as the dependent variable, and isomer structure was added to the model, coded as a dummy variable, with mutations affecting all isoforms treated as baseline. To determine whether there was any difference in age-related penetrance between the three cohorts, we examined the number of individuals carrying mutations but with no diabetes at the age of 55 years using Kaplan-Meier survival curves and χ² analysis. All statistics were carried out using the SSPS software package (SSPS, Chicago, IL).

Real-time PCR assays specific to HNF4A proved quantitative across a dynamic range of more than seven 1:2 serial dilutions. The efficiencies of detection were −3.4, −3.7, −3.7, −3.5, −3.7, and −3.3 for probe sets 1 (a, b, and c) and 2 (a, b, and c), respectively, where a gradient of −3.3 represents an optimally efficient PCR. This is a reflection of the fact that where amplification efficiencies are near 100%, there should be 3.3 cycles between each dilution point on the standard curve. The correlations between the crossing point and input template (r² value) were also good (r² = 0.99, 0.91, 0.98, 0.99, 0.98, and 0.95 for probe set 1 [a, b, and c] and 2 [a, b, and c], respectively).

In adult human liver and kidney, HNF4A expression was derived exclusively from the P1 promoter and comprised HNF4A1, HNF4A2, and HNF4A3 at 37, 46, and 17% (liver) and 42, 42, and 16% (kidney) of total HNF4A expression, respectively. In human adult pancreas and isolated islet samples, however, HNF4A expression was driven only from the P2 promoter with isoforms HNF4A7, HNF4A8, and HNF4A9 being present at 36, 42, and 22% (total pancreas) and 45, 48, and 7% (islet) (Fig. 2A and B). We also examined adult mouse islets, where only P2 isoforms HNF4A7, HNF4A8, and HNF4A9, which vary at the 3′ terminus of the gene, were expressed at 49, 47, and 4% of total HNF4A expression. A noteworthy species difference in the levels of HNF4A transcripts in liver and islets is also apparent. In our human study, levels of HNF4A mRNA are similar in liver and islets, whereas in the mouse, Hnf4a expression is much greater in the liver (39). This is in agreement with our own experiments in mouse liver and islets, where islet-specific expression of HNF4A is only ∼10% of that in the liver. In human stomach, small intestine, and colon, transcripts from both the P1 and P2 promoters were identified, although tissue-specific variation in the P1:P2 balance was noted (Fig. 2A). In adult stomach, the majority of HNF4A expression derived from the P1 promoter was present at 64% of HNF4A expression, whereas these isoforms were present only at 9% in the small intestine. Isoforms HNF4A4, HNF4A5, and HNF4A6 were not identified at significant levels in any tissue (Fig. 2A and B).

The expression levels of HNF4A transcripts differed between human adult and fetal total pancreas mRNA (Fig. 3A–C). In mature pancreas, only P2 transcripts were identified. However, analysis of human fetal pancreas (staged with respect to gestational age according to foot length) revealed the presence of P1 isoforms at ∼9 weeks post-conception (14% of total HNF4A expression; Fig. 3C). Expression was detected until the most advanced fetal pancreas was analyzed, a sample pooled from 19 and 26 weeks post-conception (Fig. 3C). The mixed derivation of fetal HNF4A isoforms made it impossible to determine the specific identity of each isoform, but the distribution of HNF4A1/HNF4A7, HNF4A2/HNF4A8, and HNF4A3/HNF4A9 isoforms, characterized by variation at the 3′ end of the HNF4A gene, showed the presence of an apparent switch in HNF4A expression between adult and fetal human pancreas (Fig. 3B and C). In the adult pancreas, HNF4A7 is present at 32% of total expression, whereas in fetal pancreas, only 8% of HNF4A expression is identified by the HNF4A1/HNF4A7 probe set (with the shorter exon 9). In the adult pancreas, HNF4A8 and HNF4A9 are present at 42 and 22%, respectively. In human fetal pancreas at 19/26 weeks post-conception, however, 77 and 15% of total HNF4A expression is accounted for by HNF4A2/HNF4A8 (with the longer exon 9) and HNF4A3/HNF4A9 (terminating in IVS8), respectively.

The median age at diagnosis for individuals with mutations in exons 2–8, which affect all isoforms, was 24 years (Fig. 4). In contrast, patients with mutations in exon 9 or 10 that do not affect HNF4A3, HNF4A6, or HNF4A9 were diagnosed at 40 years (P = 0.029). Mutations in the P2 promoter or exon 1D affect the P2 isoforms HNF4A7, HNF4A8, and HNF4A9, and the median age at diagnosis of diabetes was 31 years (P = 0.001). Because factors such as age at diagnosis may also be influenced by other factors within the families, we also carried out general estimating equation analysis to account for family relatedness. After this correction, mutations affecting only P2 isoforms caused diabetes 8.1 years later than mutations affecting all isoforms (95% CI 2.7–13.6; P = 0.010), and mutations not affecting isoforms HNF4A3, HNF4A6, and HNF4A9 caused diabetes 13.1 years later than baseline (95% CI 3.15–23.1) (P = 0.004). Kaplan-Meier analysis showed similar differences in the age-related penetrance of HNF4A mutations affecting different sets of isoforms (Fig. 5). By the age of 55 years, 90% of individuals with exon 2–8 mutations had diabetes, compared with 78% of those with P2/exon 1D mutations and only 47% of those with mutations in exons 9 or 10 (P < 0.00001).

We also correlated age of diagnosis and the position of the mutation with respect to the functional domain structure (A/B to F) of the HNF4A gene (Fig. 1). Although individuals with mutations in the F domain did present later than mutations in other regions (40 vs. 26 years; P = 0.029), this domain is also encoded entirely by exons 9 and 10. It is therefore unclear whether the effects are related to the position of the mutation within the F domain or by compensation from the unaffected P1 isoforms. Clarification of the relative contribution of these factors requires consideration of the molecular mechanisms by which the F domain exerts its effects.

The F domain interacts physically with areas of the HNF-4α protein in the D and E domains (amino acids 128–366) that are essential for its transactivational activity (31,40–44). These interactions induce a conformational change that blocks the access of cofactors such as p300 to the COOH-terminal activation function (AF)-2 motif (44,45). If the effect on age of onset is due to the position of the exon 9 and 10 mutations within the F domain, a similar effect should be noted for those regions that interact with the F domain (i.e., the D and E domains; amino acids 128–366). However, patients with mutations in amino acids 128–366 were not diagnosed later than those with mutations in other regions of the gene (24 vs. 27 years, respectively; P = 0.121). No effect of mutation type on age of diagnosis was seen (24 years for missense [n = 78] and 20 years for prematurely terminating mutations [n = 38], respectively; P = 0.139). This suggests that the observed delay in age of diagnosis of diabetes is probably attributable to compensation from unaffected isoforms rather than the position of these mutations within the F domain.

We have shown that HNF4A isoforms exhibit both temporal and spatial variation in expression, as previously reported for the HNF1A gene (29). Only the P2-derived HNF4A isoforms (HNF4A7–HNF4A9) were found in adult total pancreas RNA; the relative expression levels were 36, 42, and 22% respectively. However, in islets, HNF4A9 represented only 7% of the transcripts, suggesting that the expression of this isoform may be greater in exocrine rather than endocrine pancreas. We found no evidence for the presence of HNF4A P1 isoforms HNF4A4–HNF4A6 in any tissue. Their ubiquitous absence may indicate that they have a very restricted expression pattern or may not influence normal development or function.

HNF4A isoforms also demonstrate a developmental switch in relative expression. Transcripts derived from the P1 promoter were not observed in the adult pancreas but were present in fetal pancreas from 9 weeks until at least 19–26 weeks post-conception, comprising up to 23% of total HNF4A expression. We were unable to characterize the precise point at which P1 expression ceases, since human fetal tissues at later stages of development were not available. A developmental switch in the relative isoform expression was previously found for HNF1A, where the predominant fetal isoform is HNF1A(A), but HNF1A(B) is more abundant in the adult pancreas (29). The difference in the relative expression of the HNF1A and HNF4A isoforms implies that the pancreatic transcription factor network may be subject to isoform-specific differences in regulation.

Alternate mRNA processing of the HNF4A gene appears to be acting as a moderator of HNF4A-MODY phenotype in our study. The age at which diabetes is diagnosed and the age-related penetrance are both influenced by the position of the mutation with respect to isoform structure. The location of the mutation will determine the number and derivation of isoforms that are affected. Mutations in exons 9 or 10 are predicted to affect all isoforms except HNF4A3 and HNF4A6, which are expressed from the P1 promoter, and HNF4A9, which is expressed from the P2 promoter. Mutations in the P2 promoter or those in exon 1D are predicted to only affect P2-derived isoforms, leaving the isoforms under the control of the P1 promoter unaffected.

Our statistical analysis suggests that where some of the P1-derived isoforms escape mutation, the age of onset of diabetes may be delayed by up to 16 years. The effect is strongest for variants located in the terminal 2 exons of the HNF4A gene (median age at diagnosis 40 vs. 24 years for mutations in exons 2–8, which affect both P1 and P2 isoforms; age-related penetrance at 55 years, 43 vs. 90%), but is also apparent for mutations located in exon 1D and in the P2 promoter (median age at diagnosis 31 vs. 24 years for mutations in exons 2–8, which affect both P1 and P2 isoforms; age-related penetrance at 55 years, 78 vs. 90%). Given the small sample numbers, we cannot definitively rule out that other factors may be acting to influence age at diagnosis within our cohort, but our data still represent the largest HNF4A-MODY cohort available for study and represent a unique resource.

The reduction in effect size noted for mutations affecting the 5′ end of the HNF4A gene relative to variants affecting the 3′ terminus of the gene may reflect different modes of action for mutations located in these regions. The majority of the mutations affecting the 5′ end of the gene lie within the P2 promoter itself and are regulatory rather than coding. This raises the possibility that they may be hypomorphic and pathologically less severe. However, most of the mutations in this region lie within well-characterized binding sites for important transcription factors such as IPF-1 and HNF-1α and have been shown in in vitro transient transactivation studies and by band shift assays to profoundly affect both the binding of transcription factors and resultant HNF4A activity (27,32). The results relating to variants at the 5′ end of the gene may also be subject to some bias, since around half (27–50) of the P2 cohort are from seven families with the same C>G substitution at position −192 within the P2 promoter (5,46). The evidence for the pathogenicity of this mutation is equivocal. Although the −192 C>G mutation had only a minimal effect on HNF4A activity in in vitro studies, it did appear to disrupt the binding of an unidentified islet-specific protein to a putative regulatory motif (46). The median age at diagnosis in patients with this mutation was 45 years (5,46), which is later than for other mutations in the P2 promoter. This may be because −192 C>G, unlike the other mutations in the P2 promoter (−181 G>A and −146 T >C), does not lie within any known binding sites for pancreatic transcription factors, although it is located just outside the HNF-1α binding site.

We have demonstrated that P1-derived transcripts are present in human fetal pancreas and that mutations that do not affect these isoforms are associated with a later median age of diagnosis and also with reduced age-related penetrance. These results indicate that the differences in the expression patterns of the HNF4A isoforms between mature and fetal pancreas may play a role in moderating the activity of the pancreatic transcription factor network during β-cell development and mature function. Although further functional studies to delineate the specific roles of HNF4A isoforms are required, our analysis suggests that the HNF4A3, HNF4A6, or HNF4A9 isoforms may moderate the HNF4A diabetic phenotype. Because HNF4A6 is not expressed in either adult or fetal pancreas, and we found predominantly exocrine expression of HNF4A9 (Fig. 2B), we suggest that this effect may be attributable to the P1-derived isoform HNF4A3 expressed during development. It is possible that mutations in exons 9 and 10 could in theory disturb the expression profile of HNF4A isoforms and lead to the upregulation of HNF4A9. However, there is no evidence to support this possibility, and in the absence of islet samples from patients with such mutations, the possibility of such an occurrence is difficult to determine.

We therefore conclude that the expression in human fetal pancreas of HNF4A isoforms derived from the P1 promoter may be important in the development of the fetal β-cell and may also influence the diabetic phenotype.

This study was funded by Wellcome Trust Grant 081278/Z/06/Z and NHS R&D support. P.R.N. was supported by funding from the University of Bergen, Haukeland University Hospital, and Helse Vest. The initial design and validation of the quantitative real-time PCR assays was funded by a Diabetes U.K. small grant (BDA:RD03/0002667). L.W.H. is an RCUK Academic Research Fellow, and A.T.H. is a Wellcome Trust research leave fellow.

We thank the staff at the Department of Molecular Genetics at the Royal Devon and Exeter NHS Foundation Trust for identifying the mutations in the U.K. cohort.