Resin histology of healing incisional wounds in neonatal mouse skin reveals closure of the wound commencing within 3 hours of the lesion; by 24 hours, the epidermal wound edges have generally met and fused along much, if not all, of the length of the wound. This is true for both wild-type neonates and for PU.1 null sibs, with the only obvious differences apparent in the histology being an absence of inflammatory cells in the PU.1 null wounds (Figure 1c-j). As previously described, in situ hybridization studies using a c-fms macrophage-specific probe reveal large numbers of these cells drawn to the wound connective tissue just beneath the epidermal fusion seam at 24 hours after wounding wild-type skin, but their complete absence in PU.1 null equivalent wound sections (Figure 1k,l). The same difference, although with an earlier temporal profile, is observed if wounds of wild-type and PU.1 null skin are probed with histochemical stains that reveal neutrophil influx (data not shown). These wound dynamics lead us to believe that the neonatal mouse skin wound model provides a good opportunity to analyze the transcriptional events that regulate the various tissue-repair episodes from initial activation steps through to the 'stopping' signals that occur when the tissue defect has been filled in, both in the presence and absence of an inflammatory response.

For microarray comparison, a consistent series of horizontal and vertical incisional (criss-cross) wounds were made to the back skin of 2-day-old neonatal PU.1 null mice and their wild-type littermates. Each time-matched pair (PU.1 null and wild-type) were chosen from the same litter to reduce the possibility of differential expression 'noise' due to environmental differences. Wound tissues were harvested at either 30 minutes to identify immediate early genes, 3 hours for early tissue repair effector genes, and 12 or 24 hours to reveal later tissue repair effectors as well as inflammatory genes. Total RNA was then extracted and hybridized to Affymetrix GeneChips, and differentially expressed genes were identified by comparison of expression levels for each time point with unwounded skin samples that served as baseline controls. Genes were selected if transcript levels exceeded a twofold increase over either the unwounded baseline, or between time points, or between the wild-type and PU.1 null wounds. On the basis of these criteria, 1,001 genes were identified as wound-induced (see Additional data file 1 for an annotated database of all these genes together with full details of expression levels at all time points).

Cluster analysis with Spotfire Array Explorer 3.0 software was used to organize the 1,001 wound-induced genes into groups according to the cosine coefficient similarity measurement; this includes within a group all those genes that have a similarly shaped temporal profiles, regardless of the levels of gene expression. Nine clusters were identified in this way, and of these, seven correlated with clear episodes in the repair process. The other two had profiles that, as far as we can tell, do not correspond to any currently understood step in the repair process and so were discarded for further analysis, although they appear in our supplementary data (see Additional data file 2 for median graphs of these clusters). Of the seven clusters associated with known repair episodes, five contain one or more known genes with good functional associations to that repair episode, and this encourages us to name each cluster according to that physiological episode. This does not provide definitive proof of function for any gene in that cluster, but it gives the best opportunity to predict function, particularly for expressed sequence tags (ESTs) with no further sequence information.

Four clusters of genes have profiles that are largely independent of inflammation. Genes in these clusters are expressed with similar profile whether wounds are in wild-type skin, where there is an influx of inflammatory cells, or in PU.1 null skin, where there is none. In both these situations there is full and complete repair, and so we propose that these four clusters represent the basic repertoire of repair genes that are activated during the repair response. Figure 2 shows line graphs that display the temporal profile of the median expression levels at each time point to give a representation of that cluster and these have been termed the 'activation' (Figure 2a), 'early effector' (Figure 2b), 'late effector' (Figure 2c) and 'stop' (Figure 2d) clusters. The number of genes found in each cluster is displayed on each graph.

Three further clusters of genes represent expression profiles that correlate with the onset of inflammation and thus we consider them inflammation-associated genes. In neonatal animals, the inflammatory response is generally induced by 12 hours and is well established by 24 hours post-wounding. Two of these inflammation-associated gene clusters contain genes that are not expressed in unwounded skin or at early stages of repair; rather, they are upregulated in wild-type skin directly coincident with the onset of the wound inflammatory response but are generally not upregulated in the PU.1 null wound site at any stage. We have called these two groups of genes, the 'early inflammatory' cluster (Figure 2e) and the 'late inflammatory' cluster (Figure 2f). A third cluster does not display the standard inflammatory response profile as typified by the early and late inflammatory clusters. Rather, this cluster contains genes that are expressed at early stages of repair in both PU.1 null and wild-type mice but, whereas expression appears to increase in the wild-type wound coincident with the inflammatory response, the same genes are downregulated in the PU.1 null wound, where there is no inflammatory response; we have called this group of genes the 'inflammation-maintained' cluster (Figure 2g).

One of the most clear-cut clusters of genes is of those whose temporal expression profiles are suggestive of a transient, immediate early response to wounding. These genes show almost identical profiles, whether in the wild-type or PU.1 null situation, and thus are independent of an inflammatory response. We have named this group the activation cluster, as many will be kick-start activators of the various cell behaviors that together comprise the wound-healing process. This cluster is dominated by transcription factors and contains several well known immediate early genes, such as Egr1 (Krox 24), JunB, Myc, and I-Kappa-Bα (Nfkbia). We present a heatmap for the 90 genes in this cluster arranged with the most highly expressed at the top of the map (Figure 3a). Heatmaps provide a visual representation of temporal profiles only, and so for a small sample of these genes we also include in situ hybridization data on wounded skin sections to illustrate which cells and tissues express that particular gene. This spatial expression profile reveals expression in the in vivo setting, giving clues to the function of that gene during repair.

Krox24 (Figure 4a) has previously been shown to be transiently induced in both embryonic and adult mouse wounds 7. In situ hybridization reveals Krox24 to be expressed by those epidermal cells extending back 10-12 cell diameters from the cut edge of neonatal wounds and all the associated hair follicles within this zone also (Figure 4b).

MKP-1 (Figure 4c) is a dual-specificity phosphatase with close homology to Drosophila puckered, which has been shown genetically to be key in braking the Jun N-terminal kinase (JNK) cascade activated during morphogenetic episodes such as dorsal closure in the fly embryo 8. In situ hybridization shows that the front few rows of wound epidermis express MKP-1, although expression extends less far back from the wound edge than for Krox24 (Figure 4d). By analogy to Drosophila morphogenetic episodes, it may be that MKP-1 operates as suppressor of MAP kinase (MAPK) signaling by phosphorylation of extracellular-regulated kinases 1 and 2 (ERK1 and 2), and so may actually function as a brake on the earliest tissue movements activated at the wound site.

Expression of Fos-like antigen 1 (Fosl1, Fra1), has previously been associated with epithelial migrations during tumorigenesis but has not been analyzed in a wound-repair model 910. Fosl1 has a classic activator temporal profile (Figure 4e) and a similar spatial profile to Krox24, with high levels of expression in wound-margin epidermal cells but somewhat weaker expression in damaged hair follicles at the wound site (Figure 4f). Its close relative c-fos has previously been shown to be upregulated during repair of embryonic skin wounds 11, and in vitro studies show that blocking wound-induced fos induction may hinder cell migration 12.

Because cluster analysis allows us to group genes together that are likely to have similar functions 13, the temporal profiles of, as yet, uncharacterized ESTs in the activation cluster implicates them as having an immediate-early activator function during repair. A good example of such a gene is EST GenBank accession number AI853531 (Figure 4g), which is weakly homologous to human Mitogen-Inducible-Gene-6 (Mig-6, Gene 33). The exact function of Mig-6 remains elusive but it has been shown to interact with Cdc42, a member of the Rho family of GTPases, via the activation of stress-activated protein kinases (SAPKs) 14. In situ hybridization reveals clear expression of this gene in wound fibroblasts (Figure 4h); together with its potential Cdc42 effector status and its induction in quiescent fibroblasts upon mitogenic stimulation and expression in many human cancer cell lines 15, this suggests that Mig-6 may mediate a fibroblast migration signal. The remaining genes in the activation cluster all have very similar temporal profiles, suggesting that they too may have important roles in activating or modulating early cell behavior at the wound edge.

Two further clusters of genes have increased expression levels post-wounding in a manner that is also inflammation-independent but where expression occurs at a later time than with the activation genes. The profiles of these two clusters are temporally distinct from one another and so we have called them the early effector and late effector clusters. Between them they contain 184 genes that fit the expected profile of genes that might direct re-epithelialization and granulation tissue assembly events. The temporal profiles of all these genes can be seen by heatmap in Figure 3b (early effector cluster) and Figure 3c (late effector cluster). These two clusters contain varied types of tissue repair effectors such as tissue remodelers, genes encoding extracellular matrix (ECM) proteins, those involved in the signaling machinery and structural genes required for cell migration. Again, we provide here several examples of genes within these clusters with accompanying in situ hybridization data to provide an insight into the spatial localization of some genes in these clusters.

Map4k4, a member of the serine/threonine protein kinase family that activates the JNK and MAPK signaling pathways in response to stress signals, cytokines and growth factors 16, is a member of the early effector cluster. The temporal profile (Figure 4i) and expression of Map4k4, in both keratinocytes up to 10-12 cell diameters from the wound edge and a subset of dermal fibroblasts extending a similar distance back from the wound edge (Figure 4j), confirms the activation of this intracellular signaling cascade at sites of tissue repair. The JNK pathway has recently been shown to have a role in Paxillin regulation during fibroblast migrations triggered by in vitro scratch wounds 17, and so expression of Map4k4 is also suggestive of a cell migratory regulatory role for this signaling pathway in keratinocytes and fibroblasts during in vivo repair.

Also in the early effector cluster, retinol binding protein-1 (Rbp1), a Fabp/p2/Crbp/Crabp family retinol transporter is expressed in wound epidermal cells approximately 15 cell diameters back from the wound site (Figure 4k and 4l). This suggests a role for retinoids in re-epithelialization of the wound, and indeed, there is some evidence that these molecules can trigger epidermal proliferation via heparin-binding epidermal growth factor (HB-EGF) expression in suprabasal epidermal cells 18.

Typifying the late effector profile is Keratin 6 (K6), a classic wound-induced gene 19 (Figure 4m). K6 encodes a nonconventional keratin which is thought to facilitate the packaging up of other intermediate filaments in activated keratinocytes, so that these cells can migrate forward to re-epithelialize the wound 19. High levels of expression of K6 by the front 10-12 rows of wound-edge keratinocytes were confirmed by in situ hybridization (Figure 4n).

Interestingly, another member of the late effector cluster, the intracellular Ca²⁺-binding protein MRP8 (S100A8) is expressed in a similar temporal and spatial pattern to K6 (Figure 4o and 4p). MRP8 binds to keratin filaments as an MRP8/14 heterodimer in a Ca²⁺dependent manner 2021 and is postulated to interact with these keratin filaments and guide cytoskeletal rearrangements during tissue repair 22. The temporal and spatial coexpression of K6 with MRP8 may highlight a relationship between them and as such reveals another advantage of cluster analysis - the ability to identify potential interactions between genes and genetic pathways within the same cluster.

Not all functionally related genes cluster together, however. The heterophilic binding partner of MRP8 is MRP14, which does not appear in the same cluster but rather is expressed within the early inflammation cluster (see later), since, in addition to keratinocyte expression, it is expressed at high levels by wound leukocytes. As both the MRP8/MRP14 heterodimer and a homodimer, MRP8 is a potent chemoattractant 2223 and, interestingly, the MRP8/14 heterodimer also has an entirely different role, operating as a wound antimicrobial factor, although the MRP14 subunit seems to be responsible for this activity 24. The pleiotropic activities of MRP8/MRP14 may reflect different functions of monomeric versus complexed subunits.

At the end of the repair process many of the cell behaviors that drive repair - such as migration and proliferation - clearly need to cease as tissues re-establish approximately their pre-wound state. This will be a gradual process and yet we might expect to see such genes depressed during the repair period and becoming upregulated as wound edges meet and closure is finishing. We see a cluster of genes with exactly this profile, suggesting that some of these genes are re-expressed to control the later stages of repair. We have loosely termed this the stop cluster. Because of their known biology, several genes in this cluster make ideal candidates for players in the processes of contact inhibition and epithelial fusion that occurs as cells from the two epidermal wound fronts confront one another.

The Eph receptors and their ligands, the ephrins, have features that might make them ideal for sensing and responding to stop cues. In vitro studies show that both ligand- and receptor-bearing cells become activated upon cell-cell contact 2526, and this interaction leads to a repulsive response by receptor-expressing growth cones during the developmental wiring of the nervous system 27. Further evidence for ephrin-mediated control of epithelial sheet movement and fusion comes from studies in Caenorhabditis elegans, where Eph receptor mutants display defects in the movement of epidermal cells over neuroblasts, and in Eph knockout mice, where various morphogenetic epithelial fusions fail, leading, for example, to cleft palate and hypospadius 2829. All these results suggest that the transcriptional regulation of EphB1 revealed in the heatmaps for our stop cluster (Figure 5a) may reflect a functional role in the stopping or final fusion episodes of wound re-epithelialization.

Similarly, the expression levels of the receptor Notch also dip and rise during the repair period, and in situ hybridization studies reveal that this transcriptional regulation is also occurring within leading wound-edge epidermal cells (Figure 5b-e). Notch has exceptionally complex biology with several ligands, including Delta and Serrate, and is a widely used as a signaling cassette at various stages of embryogenesis, and has been shown to be downregulated in several invasive tumors 30. In Drosophila, Notch signaling has been implicated in the contact inhibition and fusion events that occur during dorsal closure at the end of embryogenesis (A. Martinez-Arias, personal communication), and during gut cell migratory episodes, which are also dependent on transcriptional activation of the short stop gene 31, the mammalian orthologue of which, Actin crosslinking family 7 (ACF7), is another member of our wound stop cluster.

Several other genes within the stop cluster have characteristics that indicate they may be involved in sensing contact-inhibition cues or be downstream of these signals and operate to adhere epidermal fronts together. They include genes for Plexin 3 (Plxn3), a member of the plexin family of semaphorin receptors 32, Desmocollin 3 (Dsc3), which is a cadherin component of intercellular desmosomal junctions 33 and ACF7, a cytoskeletal linker protein 34.

As with the other clusters, suggestive biology is no proof of function, and it is worth noting that several other genes with this temporal profile do not have biology suggestive of a role in these late stages of wound healing. We feel that this cluster, more than any other, can only hint at function, and definitive function testing using knockout or knockdown assays will be necessary to investigate any speculative roles in the repair process.

A comparison of those genes expressed during the repair process in wild-type versus PU.1 null mice reveals most clearly genes that are dependent on the presence of an inflammatory response at the wound site. The heatmaps for early and late inflammatory gene clusters strikingly reveal robust expression in wild-type wounds, but little or no expression in the PU.1 null mice for these genes (Figure 6). Together, the early and late inflammatory clusters comprise 169 genes that are not expressed in unwounded wild-type skin or at early stages of repair but appear to be upregulated in the wild-type wound directly, coincident with the onset of the inflammatory response. The early inflammatory cluster typically contains genes whose expression is upregulated rapidly in the wild-type, often reaching a peak by 12 hours (Figure 6a), coincident with the influx of neutrophils to the wound site. In the late inflammatory cluster, expression typically peaks a little later, at 24 hours post-wounding (Figure 6b), more suggestive of a link to the later influx of macrophages. A further 17 genes are initially expressed at both the wild-type and PU.1 wound site, but are maintained at high level only in the wild-type wound, where there is an influx of leukocytes. In PU.1 null wounds, where there is no such influx, these genes are only transiently expressed. We assume that expression of these inflammation-maintained genes (Figure 6c) is directly or indirectly dependent on signals released by inflammatory cells.

Genes that are expressed only in wild-type wounds and whose temporal expression patterns are coincident with the influx of neutrophils and/or macrophages will include those genes that are constitutively expressed by one or both of these lineages, or genes that are upregulated as part of the leukocyte activation state, or may be expressed by cells other than the invading leukocytes as a downstream consequence of host fibroblast, endothelial and muscle cells being exposed to signals from these leukocytes. We present a selection of in situ hybridization studies to illustrate each of these scenarios as revealed by very distinct classes of spatial expression pattern.

Generation and PCR genotyping of PU.1 mice has been described previously 6. Tail-tip blood smears were stained with Giesma (Sigma) for rapid identification of null individuals by the absence of neutrophils. Two-day pups received local anesthetic and full-thickness incisional wounds were made to a 1 cm × 0.5 cm area of the back skin. For the microarray study, a criss-cross network of 10 × 8 incisional wounds were made with a scalpel (Figure 1a) such that all cells within the rectangle of skin were adjacent to a cut. For in situ hybridization studies, three incisional wounds were made along the long axis of this patch of skin (Figure 1b). Subsequently, PU.1 null mice and control sibs were maintained with daily antibiotic injections until sacrifice at appropriate times. For microarray experiments, full-thickness back skin was dissected and immediately transferred, without fixation, into liquid nitrogen for subsequent RNA extraction. Wound harvesting and cryostat tissue sectioning for immunohistochemistry and in situ hybridization were performed after perfusion fixation with 4% paraformaldehyde in PBS as described 6 with tissue sections cut at 14 μm.

Total RNA was isolated from all skin samples using RNAzol (Biogenesis) according to the manufacturer's instructions with a secondary clean-up stage using an RNeasy kit (Qiagen). RNA was amplified and labeled with biotin as described previously 52.

Double-stranded cDNA was generated from 10 μg total RNA using Superscript Choice kit (Life Technologies) with a T7-poly(T) primer. Approximately 1 μg cDNA was used to generate biotinylated cRNA by in vitro transcription using Bioarray High Yield RNA Transcript Labelling kit (Enzo Diagnostics Inc). Fragmented cRNA (10 μg) was hybridized in 100 mM β-mercaptoethanol, 1 M NaCl, 20 mM EDTA, 0.01% Tween20, 0.1 mg/ml herring sperm DNA, 0.5 mg/ml acetylated BSA, 50 pM control oligonucleotide and eukaryotic hybridization controls, to Affymetrix MGU74A arrays at 45°C for 16 h. Arrays were washed using Affymetrix protocols in nonstringent buffer (6x SSPE buffer, 0.01% Tween20, 0.005% antifoam) at 25°C and stringent wash buffer (100 mM MES buffer, 0.1 M NaCl, 0.01%Tween20) at 50°C and stained with streptavidin phycoerythrin (10 μg/ml) including an antibody amplification step. Arrays were scanned using an Affymetrix confocal scanner.

The data were analyzed using Microarray Analysis Suite version 4.0 (Affymetrix). The data were scaled to a target intensity of 300. Values representing genes not expressed are unreliable and are given an absent call. All genes were subjected to a filter to identify genes with at least one present call across the full eight time points and that had on one or more occasion a greater than twofold change in gene expression levels between time points or between the wild-type and PU.1 null. The remaining genes were sorted into nine clusters using Spotfire Array Explorer 3.0 software, of which seven correlated with clear physiological episodes of repair or inflammation and we have loosely named those clusters according to these criteria.

Wound tissues were processed for resin histology with sections cut at 5 μm and stained with Toluidine Blue as previously described 6. To visualize macrophages we carried out in situ hybridization studies using the macrophage-specific c-fms probe, using the protocol outlined in 6 and see below.

Probe details are available in Additional data file 3. In situ hybridization on frozen sections was performed as previously described 53, with probe hybridization carried out in a humidity chamber at 55°C for 16 h. Expression was visualized by BCIP/NBT precipitation (Roche Biochemicals) and sections viewed on a Zeiss Axiphot microscope after mounting in Citifluor (UKC). This spatial expression information gave us a good indication of which were the expressing cell lineages in the wound site, particularly for epidermal keratinocytes (most superficial cell layer), and leukocytes (scattered individual cells in the wound granulation tissue), but without double staining with lineage-specific antibodies we cannot be definite, particularly in distinguishing specific leukocyte lineages from subpopulations of wound fibroblasts and so forth.