Genomic meta-analysis of growth factor and integrin pathways in chronic kidney transplant injury

Growth factor pathway signaling

In the first comparison of Banff 0 (n = 25) vs. All CAN/IFTA (Banff 1, 2&3; n = 59), class comparisons yielded 4683 differentially expressed probesets. The differential expression and directionality of fold changes is mentioned for each class comparison in Additional file 1: Tables S1, Additional file 2: Table S2, Additional file 3: Table S3 and Additional file 4: Table S4 for all the results discussed below. Specifically, IPA identified that these 4683 genes fall into 123 significant pathways (SD1). When we applied a Benjamini Hochberg correction for multiple testing to the pathway analysis, there were 74 pathways that were significant (p < 0.05). In this comparison HGF, VEGF, Epidermal Growth Factor (EGF), Insulin-like Growth Factor 1 (IGF-1), TGF-beta and Platelet-derived Growth Factor (PDGF) signaling pathways were all significantly differentially expressed, even with multiple testing correction.

The second comparison was Banff 0 (n = 25) vs. Banff 1(n = 14) (mild CAN/IFTA), which yielded 1430 significantly differentially expressed probesets and 84 significant pathways (SD2). HGF, PDGF, VEGF and EGF signaling pathways were differentially expressed as in the comparison to All CAN/IFTA samples. However, Fibroblast Growth Factor (FGF) was also found but the TGFβ pathway was not significantly expressed.

The third comparison, Banff 0 (n = 25) vs. Banff 2&3 (n = 29) (moderate/severe CAN/IFTA), yielded 1978 significantly differentially expressed probesets and 58 significant pathways (SD3). The EGF, FGF, HGF and IGF-1 pathways were all significant. Again, the TGFβ pathway was not significantly differentially expressed in this analysis. Finally, we compared mild to moderate/severe CAN/IFTA (Banff 1; n = 25 vs. Banff 2&3; n = 29) and revealed 1362 significantly differentially expressed genes and 44 significant pathways total (SD4). Of 44 pathways identified, there was only one growth factor pathway, IGF-1, with 7 DE genes (p = 0.05; 5/7 DE genes were upregulated in moderate/severe CAN/IFTA). However, it is important to note here that the FDR ranges for this particular class comparison ranged from 14-19%, a much less robust gene expression profile statistically than any of the other comparisons performed. Given that everything else was technically the same for all comparisons, we believe these results reflect the biology and demonstrate that the differences between early and more advanced CAN/IFTA as judged by non-quantitative histology in these studies were significantly less. It is possible that as quantitative digital histology becomes more widely available [20] that future comparisons of early to more advanced CAN/IFTA will be more robust.

MAPK dependent growth factor signaling

Many growth factor pathways are dependent on mitogen-activated protein kinase signaling (MAPK or extracellular-signal-regulated kinase; ERK). We found that LPS-stimulated MAPK, ERK/MAPK and p38 signaling were only significantly differentially expressed in the analysis of Banff 0 vs. Banff 1, 2&3 CAN/IFTA. It is worth noting that MAPK regulates signaling by TGFβ [21–23], VEGF [24–26], IGF-1 [27], and HGF [28–30]. These results further support the analysis of the growth factor pathways above suggesting the same pathways are engaged regardless of the histological severity of CAN/IFTA. The expression of the genes in the TGF, IGF-1 and HGF pathways increased in expression from mild to more severe CAN/IFTA, but the opposite trend was seen with VEGF.

Integrin pathway and expression

Transcripts for the canonical integrin signaling pathway were also significantly expressed in these latter two comparisons. There are 187 genes in the canonical integrin pathway. Interestingly, only 2 genes, PTEN and PTENP1, were significantly correlated in all three comparisons of CAN/IFTA stages suggesting a potential common signaling mechanism. Thirty-four integrin pathway genes were significantly associated with both Banff 1 and Banff 2&3 CAN/IFTA and 20 genes were only associated with Banff 2&3 (Figure 1). In conclusion, many integrin pathway molecules are highly expressed at an early stage of CAN/IFTA and continue to be expressed during the progression to more severe disease, while another set are linked to only more advanced CAN/IFTA (Table 3).

To give some context with regards to the most significant pathways associated with CAN/IFTA progression from the current study we looked at the integrin, IGF and EGF signaling pathways and their regulation with respect to the progression of CAN/IFTA. For the integrin pathway, we observed that in the Banff 0 vs. 1 comparison all differentially expressed genes in the pathway were upregulated in the Banff 1 patients. When we compared this to the expression of the same genes in the Banff 1 vs. Banff 2,3 patients some of the genes like TLN, JNK and CPN were downregulated in the Banff 2,3 patients. Similar changes were seen in both the HGF and the IGF-1 pathways with PKC and JNK (HGF pathway) being downregulated in the Banff 2,3 when compared to Banff 1, and JNK (EGF pathway) was downregulated in Banff 2,3 when compared to Banff 1, respectively. These changes suggest that some of these genes need not be overexpressed and that their downregulation in the Banff 2,3 group when compared to Banff 1 could possibly be a mechanism that promotes CAN/IFTA progression.

Proteogenomic analysis of tissue biopsies

In order to validate the results of this gene transcription profiling and functional pathway mapping at the protein level, we turned to our previously published analysis of CAN/IFTA biopsies using shotgun tandem mass spectrometry proteomics [31]. Of the 187 integrin pathway molecules, 25 (14%) of them were identified as significantly differentially expressed by proteomics (SD5) (Additional file 5). In this context it is important to note that we had already identified the whole integrin pathway and specifically integrin β1 protein as a key player in CAN/IFTA progression. In addition, proteins of the IGF-1 pathway were also significant in the progression from mild to moderate/severe CAN/IFTA (Banff 1 vs. Banff 2,3; p = 0.02) and with advanced CAN/IFTA (Banff 0 vs. Banff 2,3; p = 0.004) at the transcript level, which was similar to what we found at the protein level with the IGF-1 pathway being upregulated in the Banff 2,3 subjects, with tandem mass spectrometry proteomics. Finally, in the development of mild CAN/IFTA (Banff 0 vs. Banff 1), the VEGF and EGF pathways were significant (p = 0.02 and 0.05, respectively) by both transcript and protein.

Drug target and biomarker analysis

We used a new module in IPA based on literature and public drug target databases to determine how many of the molecules we identified by gene expression profiling in CAN/IFTA are targeted by known drugs (SD6) (Additional file 6). Of the 4109 differentially expressed genes between Banff 0 vs. All CAN/IFTA, 129 genes are targets of known drugs. In the differentially expressed growth factor pathways (HGF, VEGF, EGF, IGF-1, TGF-β and PDGF), 10 of the 78 molecules we identified are targets of known drugs. In the integrin pathway, there are only 2 genes known to be targets of drugs (ITGB5 and ITGAV) and both are differentially expressed in our data. We also used Ingenuity to identify genes that are known or potential biomarkers defined in the literature (SD7) Additional file (7). For example, of the 78 molecules associated with growth factor signaling, 34 (44%) of the genes are biomarker candidates and 368 of the total 4109 differentially expressed genes have been described as potential biomarkers in the literature in various models of disease.

The advantages of combining information from multiple existing microarray studies in a meta-analysis are a larger sample size, minimization of center-specific effects, and the application of a single data analysis pipeline. Meta-analyses also represent savings in time and costs by making comprehensive use of already available data. The present study represents a mechanism-focused, bioinformatics-driven meta-analysis of gene expression in kidney transplant biopsies with CAN/IFTA. We demonstrate that the differential activation and expression of specific growth factors, integrins and signaling pathways distinguishes the different stages of CAN/IFTA and correlate with progression of disease.

In our data, VEGF and PDGF signaling pathways were significantly upregulated in early CAN/IFTA but were not upregulated in biopsies with moderate/severe CAN/IFTA. This is consistent with an increased expression of VEGF in interstitial cells and arteries developing intimal and adventitial fibrosis in kidneys undergoing vascular rejection [32]. In contrast, only differential expression of IGF-1 signaling correlated with the progression from mild (Banff 1) to moderate/severe CAN/IFTA (Banff 2&3). Perhaps purposeful inhibition of IGF-1 signaling might be a candidate for therapy once CAN/IFTA is already present. However, these results also suggest that most of the growth factor pathways driving CAN/IFTA are present early in the course and not changing significantly as it progresses. It is worth commenting that these results also suggest caution that once patients present with early CAN/IFTA (e.g. protocol biopsy), there is a high likelihood that it will eventually progress.

Interestingly, TGFβ was only differentially expressed when comparing healthy transplants (Banff 0) to those with any grade of CAN/IFTA. However, TGFβ expression does not distinguish between mild and moderate/severe disease. Evidence for the early activation of TGFβ has been shown in a number of studies as early as 2–6 months and is often reviewed as a critical growth factor for interstitial fibrosis [33, 34]. A pediatric study showed that relatively early TGFβ expression in grafts 100 days after transplantation correlates with decreased long-term graft function and increased graft fibrosis at 3 years [35]. siRNAs directed against the TGFβ receptor improved renal fibrosis in a mouse model of interstitial nephritis [36]. Our analysis is consistent with these studies in identifying differential TGFβ pathway expression in biopsies with CAN/IFTA compared to healthy transplant biopsies. However, the fact that there is no further increase in TGFβ gene expression despite increasing severity of disease suggests that other pathways mediate progressive fibrosis and tubular atrophy.

The HGF and FGF signaling pathways were significantly upregulated in early CAN/IFTA, but unlike TGFβ were further upregulated in biopsies of moderate/severe CAN/IFTA. HGF, a member of the FGF family, is known to mediate the repair and maintenance of the kidney epithelium. HGF is nephron-protective [37] and suppresses interstitial fibrosis in a rat model of obstructive nephropathy [38]. Exogenous administration of HGF has been shown in a rat model of CAN to prevent the progression of IFTA [39]. Thus, one possibility for our results in these human patients is that up-regulation of the intrinsic HGF pathway acts as a compensatory mechanism protecting the progression of chronic injury and fibrosis. It adds caution to the tendency of advocating blockade of any pathway found in any pathological condition.

FGF is known to stimulate endothelial cell proliferation, degrade ECM and interact with VEGF [40]. In situ hybridization and immunohistochemical analysis showed that mRNA for FGF-1 and its high-affinity receptors were increased in the tubular epithelium, inflammatory cell infiltrates, and neovascular structures of kidney transplant patients who underwent nephrectomy after graft loss for what the authors called chronic rejection [41]. Our results are consistent with this role for FGF-1 in biopsy-proven CAN/IFTA.

As already noted, VEGF and PDGF signaling pathways were significantly upregulated with mild CAN/IFTA, but not further upregulated with moderate/severe CAN/IFTA. In a different set of transplant biopsies done several years ago, we also demonstrated the up-regulation of PDGF with CAN/IFTA [42]. Thus, it may be that VEGF and PDGF signaling is important early in the establishment of chronic allograft injury, perhaps by amplifying the injury driven by FGF-1.

The MAPK-dependent growth factor signaling pathways, ERK and p38, were significantly differentially expressed in our study between well functioning transplants vs. any grade of CAN/IFTA. In rat models, the inhibition of p38 MAPK by treatment with a specific p38 MAPK inhibitor resulted in reduced CAN/IFTA with preserved renal function and survival [43, 44].

We also show that specific integrins are upregulated in the progression of CAN/IFTA (Table 2) and these results correlate with the parallel up-regulation of the canonical integrin signaling pathway (Figure 1). The integrin subunits β1, β6 and α6 were consistently and significantly differentially expressed in all comparisons. An earlier study in rats demonstrated that blockade of interactions between α4β1integrin and fibronectin prevents the development of chronic rejection in cardiac allografts [45]. However, we note that our kidney results do not demonstrate differential expression of the α4 chain. One explanation is that the β1 subunit regulates the assembly of the α4β1integrin in kidneys. A study of integrin expression in allograft rejection showed that αvβ6 was involved in the re-epithelialization of healing isografts in an animal model of chronic progressive lung allograft rejection [46]. In two other studies, mice with bleomycin-induced pulmonary fibrosis were treated with an anti-αvβ6 monoclonal antibody [47, 48]. Treated animals showed reduced fibrosis and these results were confirmed using transgenic and knock-out animals. In a published meta-analysis of human kidney transplant biopsies, 309 genes were identified as associated with CAN/IFTA [14]. We identified 47 genes from this set in the present analysis including the α6 chain. Thus, these many lines of evidence converge to identify roles for several specific growth factors and integrin adhesion pathways in chronic kidney transplant injury.

Our analysis of the correlation of the genomic findings with our previously published proteomic work clearly identified the integrin pathway (Figure 1) and the IGF-1 pathway as being differentially regulated and as pathways that specifically affect CAN/IFTA progression. This is very relevant because both these pathways are validated both at the mRNA as well as the protein level and could potentially be the highest priority pathways to address with animal models to study specific mechanisms of CAN/IFTA.

Finally, we used a new module in Ingenuity Pathway Analysis to search for drug targets in our differentially expressed genes and pathways (SD6). For example, in the context of the integrins, there are three drugs that target the integrin αv subunit, two antibodies and one cyclic RGD pentapeptide and the pentapeptide also targets another differentially expressed integrin subunit, β5 [49]. Ten of 78 molecules involved in known growth factor signaling pathways (HGF, VEGF, EGF, IGF-1, TGF-β, PDGF) were targets of over 30 different drugs. Many of these drugs are already FDA-approved while others are in various phases of clinical trials for cancer, hypertension and diabetes, including specific kinase inhibitors, receptor signaling blockers and monoclonal antibodies. The potential of using bioinformatic tools to integrate public data with clinical gene profiling results and identify drug targets or candidate biomarkers is still largely untested. However, two recent studies from Stanford used a similar approach to identify drug targets for cancer, one of which they validated, or search for transplant biomarkers [50, 51].

We would also like to draw the attention of the readers to some of the limitations of our study. First, even though we used 84 samples to perform a meta-analysis and despite the fact that we were adequately powered to detect gene expression changes greater than two-fold, the findings of the current study need to be validated in a larger cohort of samples, especially of the different subtypes of CAN before it can be generalized. Similarly, we cannot discount the effect of different immunosuppressive drug regimens and levels on the gene expression, though all the studied patients were on a primary calcineurin inhibitor protocol. The correlations between transcript levels and protein levels have been shown to be relatively poor in earlier studies[52, 53]; however several recent studies have shown that these correlations are a lot better than previously shown due to the technical advancements in protein detection by the latest mass spectrometry technologies[54, 55]. Our current study used tandem mass spectrometry proteomics data from an earlier study of ours using liquid chromatography coupled with an ion trap instrument (MudPIT protocol for LC/MS/MS) [31]. Newer technologies have significantly enhanced mass detection accuracies and targeted proteomics using alternative mass spectrometry technologies such as multiple single reaction monitoring (MSRM) and triple quadropole TOF instruments will now allow investigators to more accurately identify and even quantify protein level changes for better correlations with gene expression levels profiled in the same samples. The importance of the present work is to establish a clear set of candidate genes and functional pathways that would merit the next stage of experimental efforts to validate these with proteomics.

Microarray analysis of tissue biopsies

Proteogenomic analysis of tissue biopsies

All proteomic data comparisons were made using our previously published study that revealed about 1400 proteins with unique expression profiles tracing the progression from normal transplant biopsies to increasingly severe grades of CAN/IFTA [31].

Consent

Written informed consent was obtained from all the patients in this study for publication of this manuscript.