Gene expression profiling of mammary gland development reveals putative roles for death receptors and immune mediators in post-lactational regression

Six-week-old C57/Bl/6 mice were obtained from Harlan Laboratories (Bicester, UK). At 8 weeks females were either mated or culled for virgin mammary glands. Postpartum, a minimum of six pups per suckling female was ensured by cross-fostering where appropriate. Females used in involution time points were force weaned at 10 days of lactation. Lactating and involuting mammary glands were monitored daily for signs of localized inflammation or mastitis. Two out of 90 animals were excluded from the study for this reason. The study was approved by the local ethics committee and complies with the Helsinki Declaration.

A total of 12 stages of adult mouse mammary gland development were selected for this study: 8-week-old virgin; 5 days, 10 days and 15 days of gestation (in which day 1 was the first day postcoitum); 0 days (first day postpartum), 5 days and 10 days of lactation; and 12, 24, 48, 72 and 96 hours after forced weaning. All mammary glands were harvested between 11:00 and 13:00 to limit circadian effects. A total of six animals were used per time point (three per hybridization). In each case a single abdominal gland was removed following excision of lymph nodes and immediately frozen in liquid nitrogen. RNA was extracted from frozen tissue using Trizol reagent (Invitrogen, Paisley, UK) followed by additional column purification (Rneasy; Qiagen, Crawley, UK). Briefly, three samples of mammary tissue from the same time point were ground under liquid nitrogen, and 20 mg of each was pooled into 1.5 ml Trizol reagent. Resuspended RNA samples were then passed through purification columns according to the manufacturer's instructions and RNA integrity was monitored with Lab-on-a-chip (Agilent, West Lothian, UK) before use in microarray analysis.

RNA 5 μg was labelled according to manufacturers protocols (Affymetrix, High Wycombe, UK) and hybridized to Affymetrix MGU74ver2a chips representing 12 488 transcripts or 8618 genes. Computation of expression values was performed using the perfect match/mismatch model implemented in dChip [20].

Both data sets were clustered (K means) according to expression pattern similarity across all 12 conditions (complete time course) or seven conditions (involution time course) using standard correlation. Nine K mean classifications were performed for each data set, starting with 10 clusters and increasing by five incrementally to 50 clusters. The optimal number of clusters was determined empirically based on highest observed variability and redundancy between similar clusters.

All genes were annotated according to known function using the Gene Ontology Consortium categories [21]: biological process, cellular component and molecular function. Onto-Express [22] was used to determine whether clusters of genes with similar expression profiles were enriched in specific GO functional categories. Based on the genes present on the GeneChip, Onto-Express calculated the expected number of occurrences of each functional category in each cluster. The probability that each functional category was over-represented in a cluster was derived using a binomial model and the P values were corrected for multiplicity using the Bonferroni method. Thus, Onto-Express provided information about the statistical significance of each of the pathways and categories represented by the genes in each cluster (these data are available online as Additional files 1 and 2).

All genelists, including expression profile clusters, ontological definitions and all raw data .cel files, fulfilling MIAME criteria, were uploaded to ArrayExpress http://​www.​ebi.​ac.​uk/​arrayexpress/​ and are also available through GeNet http://​genet.​hgmp.​mrc.​ac.​uk:​8080/​servlet/​GeNet for download and online analysis. (For information on accessing data through GeNet or from Breast Cancer Research see our website: http://​www.​path.​cam.​ac.​uk/​~madgroup/​). Gene lists use Affymetrix probe_ID and accession number as transcript identifiers. In the body of the text and figures, gene/transcripts are named according to the Mouse Genome Database [23].

It was expected that a proportion of the 6797 differentially expressed transcripts would be specific to involution, whereas others may have additional roles at earlier stages in the pregnancy cycle. In order to identify these subsets of involution related genes, the expression patterns of all differentially expressed transcripts were compared and subsequently grouped according to similarity into 35 gene profile clusters. Some clusters exhibited superficially similar patterns (Fig. 1). In order to simplify the classification of expression patterns observed in the time course, these similar profiles were pooled, resulting in 11 broadly distinct expression profiles (Fig. 1a).

These 11 clusters represented maximal gene expression at different stages during the pregnancy cycle. Thus, one cluster of 651 transcripts (5% of all transcripts on the array) were maximally expressed in virgin mammary gland (V_SL). Two clusters with a total of 2317 transcripts (19%) were maximally expressed in gestation (G and G_SL) and three clusters (529 transcripts, 4%) were maximally induced during lactation (L_T, L and L_G). Four clusters (3113 transcripts, 25%) exhibited maximum expression during involution (I_L, I_T, I_P and I_G) whereas one cluster included 187 transcripts (1%) that were uniformly expressed throughout the pregnancy cycle (PC). Further details on the nomenclature of these combined clusters are provided in the legend to Fig. 1. Genelists for each combined cluster and the 35 primary clusters are available online (see Method section, above).

Six clusters consisted of transcripts that were either maximally expressed during involution (I_L, I_T, I_P and I_G) or specifically suppressed following weaning (L and L_G), suggesting that a proportion of these genes may play important roles in regression of the mammary gland. These six clusters consisted of 3463 transcripts, or 28% of all transcripts on the array. Only one group of 518 genes exhibited involution-specific expression, all of which were transiently activated within 24 hours after weaning (I_T). It was expected that a proportion of these represent genes that may contribute to the initial phase of cell death characteristic of the reversible first stage of involution. All other involution related clusters exhibited elevated expression at earlier stages in the pregnancy cycle. For example, one group of 236 involution related genes (I_P) exhibited an additional peak of expression specifically at parturition. This suggested that many transcriptionally regulated processes in involution were common to gestation (I_G), parturition (I_P) or lactation (I_L).

Of the 6747 differentially expressed transcripts, 3273 (49%) encoded proteins with known function. In order to establish what proportion of the transcripts in each cluster from Fig. 1a shared common biological functions, transcripts with known function were assigned to categories as defined by the Gene Ontology Consortium http://​www.​geneontology.​org, and statistically significant associations between gene function and each cluster were identified using Onto-Express [22]. The significant associations (corrected P < 0.05) are listed in Table 1. Further details on the distribution of functionally related genes between these clusters are available in Additional file 4.

Unexpectedly, similar proportions of apoptosis related genes were expressed in each of the 11 clusters; the only significant correlation being with the I_L cluster, in which maximal gene expression occurred at 12 hours of involution. However, a number of other biological processes were significantly linked to particular stages of mammary development, and were consistent with our current understanding of the physiology of the mammary gland. Thus, during gestation there was a significant increase in the expression of cell cycle regulatory genes (G, G_SL and I_G) and a concomitant increase in nucleic acid and macromolecular synthesis ('biogenesis' and 'metabolism' in Table 1). By lactation, the proportion of cell cycle genes had diminished, to be replaced by an increase in the number of genes involved in development and differentiated function, including fatty acid biosynthesis and other metabolic processes (L_G, L and I_L). These three clusters should therefore contain genes that are important for terminal differentiation and transition to a secretory phenotype (genelists for each cluster are available online; see Method section, above).

Morphogenesis and tissue remodelling are characteristic features of the mammary gland both during pregnancy and late in postlactational regression. We observed significant correlations between morphogenesis and three clusters with broadly similar expression profiles (L_G, L and I_L), each being characterized by increased gene expression during gestation and lactation followed by a marked and progressive decline at the lactation–involution boundary. A significant proportion of remodelling-related genes were also observed in the V_SL cluster, which exhibited an inverse pattern to the three profiles described above. Because the transcripts in the V_SL cluster were, by definition, distinct from those in the L_G, L and I_L groups, it is possible that the former may contain remodelling related genes that contribute to mammary regression whereas the latter group of three clusters contains genes that are involved in invasion and remodelling of the fat pad during pregnancy.

The major differences between these two groups of genes were the high number of cell adhesion molecules (CAMs; e.g. Ceacam-10 and -11, integrins α5 and β3, Vcam1, Ncam1, Pecam and Cadherin-8) and bone morphogenic protein (Bmp5 and Bmp7) genes in clusters L_G, L and I_L compared with the relatively high proportion of matrix components (e.g. laminin-α2, -α4 and -β2; procollagen types Iα1, Iα2, IVα1, IVα2, Vα1, VIα1, VIα2, VIα3, XIVα1 and XV; fibronectin-1 and decorin), matrix metalloproteinases and inhibitors (Mmp9, Mmp14, Timp3) and fibroblast growth factors (Fgf1, Fgf2) in the V_SL cluster.

A strong statistical relationship existed between involution and immune-related genes (Table 1 and Fig. 2). Of the four clusters exhibiting maximal gene expression during involution, three were significantly associated with inflammation (I_T), the acute phase response (I_P and I_L), or humoral immunity (I_P). Of the remaining eight clusters, V_SL, whose transcripts exhibited a modest increase during involution (and were linked with remodelling), was associated with innate cellular defence (Fig. 2). Immune genes were most prevalent in the I_P cluster, in which transcripts were transiently expressed at parturition and maximally expressed by 48 hours of involution (Fig. 2). Of the 57 immune-related genes present in the I_P cluster (24% of known genes in the cluster), 43 (18%) were immunoglobulin genes, seven (3%) were acute phase protein genes and seven (3%) were inflammatory mediators.

We have previously shown that the IL-6 family cytokine leukaemia inhibitory factor (LIF) induced phosphorylation (and activation) of Stat3 in mammary gland during involution, and that this transcriptional signalling pathway was important for the induction of apoptosis in the early involuting gland [17, 24]. Stat3 is also a recognised mediator of the acute phase response, and because acute phase related genes exhibited a parturition/involution expression profile (Fig. 2) we assessed the relationship between Stat3 activity and acute phase gene expression.

Acute phase proteins are classified according to the transcriptional signalling pathways that regulate their expression [25]. Thus, class I acute phase genes are activated by cooperative binding of CAAT-enhancer binding protein (C/ebp)β, C/ebpδ and/or nuclear factor (NF)-κB (and in some cases also Stat3) to specific sites in their promoters. Class II genes are activated by Stat3 alone. Class I genes were induced at parturition and involution (similar to the I_P profile). Class II genes were transiently, and specifically, expressed at the onset of involution (similar to the I_T profile). We previously described Stat3 and NF-κB activities in the mammary pregnancy cycle [24, 26], whereas Cebp gene expression in mammary gland has also been described [27]. We compared the expression of these transcription factors and found that Stat3 and both Cebp genes in addition to phosphorylated Stat3 DNA binding activity [17] correlated with class I acute phase gene expression (Fig. 3).

In order to define patterns of gene expression pertinent to the lactation–involution transition, irrespective of expression in pregnancy, differentially expressed genes were reclustered using a truncated time course involving lactation and involution time points only (Fig. 4). A total of 4103 transcripts (33% of all transcripts) exhibited significant changes in expression across the seven time points, namely 5 days and 10 days of lactation, and 12, 24, 48, 72 and 96 hours of involution. These profiles were sorted according to similarity into 15 clusters (Fig. 4b). Similar clusters were then combined to produce seven basic profiles that differed according to the kinetics of gene induction or suppression during lactation and involution (Fig. 4a). Thus, four patterns of activation were observed: a rapid but transient increase in gene expression 12 hours after weaning (Inv1); a rapid activation, which was maximal by 12 hours and sustained for up to 4 days (Inv2); a slightly delayed induction, maximal by 24/48 hours, and sustained for at least 96 hours after weaning; and a gradual increase in expression, which peaked at or beyond 96 hours (Inv4). Similarly, three patterns of suppression were observed: a transient decrease in transcript levels that mirrored the pattern observed in Inv1 (Lac1); a rapid loss of expression, reaching baseline levels at 12/24 hours and remaining low for up to 96 hours (Lac2); and a delayed reduction in expression, originating from a peak in expression at 12 hours involution and reaching a minimum beyond 96 hours (Lac3).

These patterns inferred three phases of gene expression during involution: rapid changes within the first 12 hours (Inv1, Inv2, Lac1, Lac2), delayed expression by 24–48 hours (Inv3), and prolonged changes extending beyond 4 days (Inv4, Lac3).

Consistent with our analysis of the complete 12-point time course, we found that involution related gene expression was associated with an overall increase in immune gene expression (Table 2 and Fig. 5; also see Additional file 5). There was an initial transient increase in proinflammatory genes (Inv1). However, the association was most pronounced with transcripts exhibiting delayed kinetics (Inv4), in which known genes involved in innate immunity, antimicrobial defence and inflammation were all preferentially expressed. The delayed immunological changes also coincided with a significant increase in apoptosis markers (see below) and factors that affect tissue architecture, including genes encoding structural proteins (decorin, laminins, Nid1, collagen 1α1, 1α2, 3α1 and 4α2) and matrix associated proteins (Mmp3, Mmp12, Gelsolin, uPa and Adamts1), thus corroborating previous reports of the role of matrix proteases in remodelling of the regressing mammary gland [2]. Lactation specific expression, as expected, was associated with downregulation of differentiation and metabolic processes after weaning (Lac2 and Lac3; Table 2).

In order to gain further insight into the immune responses associated with mammary involution, we listed the cytokines, chemokines and immune-related genes associated with each involution profile from Fig. 4 (Table 3). This confirmed the two phases of immune-related gene expression illustrated by Fig. 5: an initial transient expression of proinflammatory cytokines and TNF superfamily genes followed by delayed expression of monocyte and lymphoid chemokines and immunoglobulin genes.

Thus, an initial burst of cytokine and cytokine receptor expression was observed within 12 hours of forced weaning (Inv1; Table 3), which included the acute-phase mediators LIF receptor (Lifr), IL-11 (Il11) and the neutrophil chemoattractant Gro1 (Cxcl1), and the proinflammatory mediators IL-1a, IL-1b and IL-13 (Il1a, Il1b and Il13). This was accompanied by a transient decrease in other cytokine receptors (including Il11r) and an inhibitor of IL-1 receptor signalling (Il1rn), suggesting that specific proinflammatory signals may be suppressed at the transcriptional level.

One expected outcome of the transient cytokine expression was an infiltration of inflammatory cells, in particular, neutrophils due to the expression of Cxcl1. Analysis of immune cell specific antigen expression demonstrated a marked increase in the monocyte/macrophage markers Lrp, Csf1r and Cd14 (Table 3 and Fig. 6a), but no evidence for an accompanying increase in neutrophils or eosinophils (polymorphonuclear leucocyte markers Fut4, Fcgr3, Caecam1, Il8rb and Tnfrsf5 were not induced during involution; data not shown). This is consistent with previous observations of the immune cell complement in regressing murine mammary gland [28] and supports the hypothesis that neutrophil numbers are suppressed in involuting tissues in order to prevent subsequent tissue damage [28, 29]. Interestingly, the gene for oncostatin M receptor (Osmr) was upregulated in involution (Fig. 6b). In models of inflammation oncostatin M is produced locally to prevent proinflammatory cytokines such as IL-8 from recruiting neutrophils, thus attenuating the inflammatory response [30]. We also confirm a previous report that uterocalin (Lcn2) and Sgp2, a subunit of clusterin, are upregulated during involution and parturition [19], because their expression correlates with the I_P profile (Fig. 6b). Uterocalin, an acute-phase response gene that is regulated by Stat3 [31, 32], prevents neutrophil accumulation possibly through targeted destruction of neutrophils [29], whereas clusterin protects cells from the damaging effects of neutrophils [33, 34]. Furthermore, secretory leukocyte protease inhibitor (Slpi), a potent inhibitor of neutrophil proteases [35], and a Stat3-induced gene (Boland M and coworkers, unpublished data) was also markedly upregulated during involution (Fig. 6b). Thus, although the transcriptional effects of the acute-phase mediators LIF–Stat3 and C/ebpβ/δ are clearly evident (Fig. 3), the immunological outcome at the cellular level is likely to be minimised by coexpression of anti-inflammatory factors such as these.

Following the initial 'spike' in inflammatory cytokine expression, six different chemokine genes were slowly induced over the 4-day period of involution (Inv4; Table 3). These included chemoattractants for monocytes and macrophages (Ccl6, Ccl7, Ccl8 and Cxcl14) and lymphocytes (Ccl6, Ccl21a, Cxcl9 and Cxcl14). This chemokine induction was accompanied by a marked increase in immunoglobulin gene expression (Fig. 5, Table 3). Consistent with these observations, the monocyte/macrophage markers Lrp1, Csf1r and Cd14, and the lymphocyte antigens Tnfrsf9, H2-M1 and Tnfrsf4 were increased during involution (Table 3 and Fig. 6a).

In addition to the immunoglobulins, genes that encode a number of innate soluble defence factors were highly expressed in involution (Table 3 and Fig. 6c). These soluble factors possess antimicrobial properties that are secreted to protect against mastitis during the vulnerable period of milk stasis [36]. Thus, the iron chelator lactoferrin (Ltf), which is found at high levels in the milk of humans and cattle [37, 38], is induced with kinetics similar to those of the antibody transcripts described above (Fig. 5). Complement components are prevalent at all stages during involution (Table 3). These factors are found at high levels in mastitic milk and in normal bovine mammary glands during involution, and are thought to play a dual role in opsonization and lysis of bacteria and as a proinflammatory mediator [39, 40]. Cd14 was significantly upregulated in involution and was also highly expressed during parturition (Fig. 6c). This antigen is expressed on bovine mammary polymorphonuclear neutrophils and macrophages [41] and plays an integral role in suppressing bacterial infections in response to endotoxin. Furthermore, Lbp [42] and Ly86 [43], which in conjunction with Cd14 augment responses to lipopolysaccharide (LPS), were also upregulated. The significant regulation of these genes in uninfected mammary gland confirms a role for soluble defence factors in normal development of mouse mammary gland.

Involution is characterized by extensive apoptosis of epithelial cells, which occurs both before and during the tissue remodelling process. By subdividing the first 4 days of involution into three discrete phases of gene expression, we demonstrated a link between the expression of apoptosis-related genes and the delayed onset of extracellular matrix degradation and alveolar collapse at 48 hours (Inv3 and Inv4; Table 2). However, we were unable to identify a significant increase in the number of apoptosis genes expressed at the onset of involution. One explanation for this was that specific combinations of cell death genes, rather than an increase in the number of genes activated, influenced apoptosis at the onset of involution. In order to address this, all cell death related genes found to be regulated during involution were grouped according to their lactation/involution expression pattern (Table 4). Regulators (Bax, BclX, Mcl1, mXIAP and Apaf-1) and effectors (caspases) of classical apoptosis and immune cell-mediated killing (granzyme A, granzyme B, Fadd, Fas, Fas ligand) were regulated with different kinetics during involution and were relatively evenly distributed across the four involution clusters (Table 4).

However, we observed a strong bias in the expression of extrinsic (death receptor) and intrinsic (mitochondrial) apoptosis genes, in which extrinsic-related components were generally induced within the first 24 hours of involution whereas intrinsic genes, with the exception of Hrk and Diva, were not maximally transcribed until at least 96 hours.

Thus, six ligands of the TNF superfamily (Tnf, Tnfsf4, Tnfsf6, Tnfsf7, Tnfsf10 and Tnfsf12) were specifically induced in involution and all were transiently activated at 12 hours after weaning (Inv1; Fig. 6d). Four of these ligands (death receptor ligands: Fas ligand, TNF-α, TWEAK and TRAIL) are responsible for activating the five principal mammalian death receptor (extrinsic) apoptosis pathways through their cognate receptors Fas, TNFR-1, TNFR-2, DR3 and DR4. Two of these receptors (Fas and Tnfrsf1a) were also induced, and maximally coexpressed, within 24 hours of weaning (Inv3; Table 4). This correlates with previous studies in mammary gland of two of these ligand–receptor pairs [44, 45]. NF-κB activity also correlated with the rapid activation of these TNF superfamily ligands. DNA binding activity of this transcription factor is markedly upregulated within 3 hours of forced involution and is suggested to promote survival of a subpopulation of mammary epithelial cells [26]. Thus Nfkb2 and activators of NF-κB (Tnfsf12, Bcl10, Tnfrsf11a, Traf6 and Tnfsf10) were upregulated, whereas an inhibitor of NF-κB function (Tnip1) was inhibited during the initial 24 hours following forced weaning, when NF-κB activity was induced [26]. Previous studies demonstrating the direct activation of NF-κB by TNF superfamily ligands in mammary epithelial cells [26, 46, 47] therefore suggests that the observed transient increase in these ligands may be functionally relevant.

In contrast to the rapid response of extrinsic factors, intrinsic apoptosis genes were predominantly (but not exclusively) expressed at least 24 hours after forced weaning. Proapoptotic and antiapoptotic members of the Bcl-2 family (Bax, Mcl1 and Bcl2l) were induced together with the downstream effector of the intrinsic pathway Apaf1. Although the precise mechanism(s) of cell death in mammary gland have yet to be established, the trends in apoptosis gene expression identified here hint that extrinsic (death receptor) and intrinsic death programmes may predominate at different times. A marked induction of Igfbp5 complemented the delayed induction of intrinsic apoptosis genes (Fig. 6d). The kinetics of this potent inducer of mammary apoptosis and involution [48] suggests that it probably contributes to the stimulation of intrinsic cell death at this time.

Apoptotic bodies initially shed into the lumen and the residual dying cells left in the epithelium are believed to be cleared from the mammary gland by nonprofessional phagocytes and interstitial macrophages, respectively [28]. Receptors and ligands that mediate recognition and engulfment of apoptotic cells include members that are common to both types of phagocytic cell (e.g. Cd36, Lrp1 and Mfge8) and those specific to professional macrophages (e.g. Cd14 and Cd68). The genes for each of these factors (except Cd36, which exhibits an initial downregulation before a modest increase by 96 hours) are induced with similar kinetics (Fig. 6e). This is indicative of the increasing demands placed on the mammary gland to clear dead/dying cells during regression, and also probably reflects the increasing numbers of macrophages occupying the gland during the first 4 days of involution.

Our data also provide provisional evidence for the existence of a two-phase apoptotic process in involution and which also coincides with the two phases previously described [2]. While the number of apoptosis genes was similar in each involution cluster, there was a pronounced bias in the distribution of extrinsic and intrinsic apoptosis genes. The initial spike of activity consisted of four death receptor ligands, namely Tnf (TNF-α) Tnfsf6 (Fas ligand), Tnfsf10 (TRAIL) and Tnfsf12 (TWEAK). These ligands bind to the principal death receptors Fas, TNFR-1, TNFR-2, DR3 and DR4, which initiate apoptosis in cell mediated immune cell death and in a variety of physiological and pathological situations. These data confirm previous observations that Fas and Fas ligand are upregulated within a day of involution in mouse mammary gland, and that the deletion of these genes results in delayed regression and loss of apoptosis [44]. Deregulation of Fas signalling through aberrant Akt and NF-κB activity is believed to play a role in the immune escape of breast cancers and in resistance to chemotherapy [49]. TNF-α also induces NF-κB signalling to prevent apoptosis in a subpopulation of cultured mammary epithelial cells [26, 47], and has been shown to be downregulated late in involution (day 7); however, it has not been studied at earlier time points [50]. Furthermore, elevated IL-10 dependent TRAIL expression (and its receptor DR4) has previously been demonstrated in involution [45].

In contrast to the transient activation of death receptor ligands, components of mitochondrial (or intrinsic) apoptosis exhibit sustained, often delayed induction. Thus the caspase 9 effector, Apaf1, and Bax were induced with delayed kinetics. Costimulation of the survival factors Mcl1 and Bcl2l (BclXl) suggests that the transcriptional regulation of Bax may be counterbalanced, and therefore controlled, by expression of Bcl-2 homologues. Suppression of BclX, for example, promotes death during mammary involution [51], whereas delayed induction of Bax has previously been demonstrated [52]. Regulation of Bcl homologues was not exclusively restricted to late stage involution. Hrk and Boo/Diva were differentially expressed within 24 hours. Hrk is a proapoptotic factor that is transcriptionally suppressed by progesterone signals in breast cancer cell lines [53]. The function of Boo/Diva, which has only previously been described in mouse ovary, is poorly defined [54]. Cross-talk between death receptor pathways and Bcl-2 regulatory factors is well established [55] and may explain the early activation of these two Bcl homologues. Probable upstream initiators of a putative delayed apoptotic mechanism include the loss of survival signals, such as insulin-like growth factor (IGF)-1 [48, 56], and the destruction of extracellular matrix–integrin contacts leading to an anoikis-type death [6, 57]. The dramatic induction of the proapoptotic mammary protein IGF-binding protein 5 proposed previously [48] coincides with the onset of second phase involution (Fig. 6d, Table 4) and the upregulation of metalloproteinases [2, 6]. Both IGF-1 and integrins signal through Akt, which was also downregulated in involution (Table 4). Loss of this signal could then induce Bax [57]. Our data suggest that apoptosis is induced in involution following the specific transcriptional activation of a subset of proapoptotic genes encoding death receptors, and that delayed expression of inhibitors of cell survival (e.g. IGF-binding protein 5) induce alternate apoptosis pathways during the second phase of involution. This hypothesis warrants further investigation.

Integral to the apoptotic process is the phagocytic clearance of apoptotic bodies by macrophages. It has been suggested that monocytic macrophages are recruited to the mammary gland to supercede phagocytic epithelial cells in the clearance of milk and dead cells [28]. We have demonstrated monocyte specific (Cd14 and Cd68) and nonspecific phagocyte receptors during involution, supporting a role for monocytic phagocytes in mammary involution.

Measurements of the cellular content in milk from involuting glands of sheep, pigs and cattle indicate an early rise in polymorphonuclear neutrophils, a subsequent increase in macrophages and low numbers of eosinophils, B cells and T cells [7, 28]. However, the immune cell complement of human and rodent mammary gland is poorly defined. It is believed that these phagocytes are recruited to the gland to perform several functions in phagocytic clearance and cellular immunity.

We have described the sequential activation of cytokine and chemokine genes in involution. The effect of these would be to attract monocytes and lymphocytes to the gland, and we identified an increase in a number of immune cell markers, supporting this hypothesis. For example, increased expression of the Csf-1 receptor, c-fms (Csf1r; Fig. 6a), which is restricted to macrophages of the monocyte lineage in mammary gland [58] and is required for appropriate branching morphogenesis in pregnancy [59], confirmed that macrophages are critical for normal development [58]. A role for these cells in phagocytic clearance during involution is emphasized by the concomitant increase in two receptors (Cd14 and Cd91) and the milk factor Mgfe8, which mediate recognition of apoptotic cells by macrophages (Fig. 6c, 6e).

Recruitment of cells of the monocytic lineage comprises an inflammatory response. However, in the absence of a persistent bacterial infection, this does not occur in the regressing mammary gland, even though inflammatory cytokines and downstream mediators are activated, presumably to perform additional roles during involution. During chronic inflammatory diseases such as rheumatoid arthritis or inflammatory bowel disease, uncontrolled accumulation of neutrophils yield aberrantly elevated levels of inflammatory compounds, which become major contributors to tissue damage [60, 61] Thus, regulation of neutrophil recruitment into inflammatory sites and their clearance are critical processes that ensure effective host defence without tissue injury. In mammary gland, the LIF–Stat3 signaling pathway is known to be crucial for appropriate regression during early involution by initiating epithelial apoptosis [17, 24]. With the concomitant downregulation of IL-11 receptor (Table 3) and lack of a significant increase in IL-6 or its receptor (data not shown), it is also likely to be the predominant protagonist of acute phase signalling. This is confirmed by previous observations that C/ebpβ and C/ebpδ are both transcriptional targets of Stat3 in vivo [24] (Boland M, unpublished data) and that the profiles of Class I acute phase genes correlated precisely with phosphorylated Stat3 activity [24] and Stat3, C/ebpβ and C/ebpδ expression (Fig. 3). Class II acute-phase response genes exhibit specific and transient expression in involution. It is thought that these genes respond transiently to proinflammatory signals [25] and therefore may respond in a similar manner to the rapid activation of Stat3 at the onset of involution. Why they are not induced at parturition is unclear, although a molecular switch that determines the specificity of signalling pathways downstream of LIF has been proposed [24].

Thus, inhibitory mechanism(s) must exist to prevent the accumulation of potentially damaging cellular infiltrates in response to LIF. In a previous study of involution in uterus and mammary gland, the acute phase protein uterocalin (Lcn2) was shown to be induced [29]. It was proposed that one role of this protein in involuting mammary gland and uterus was to eliminate infiltrating neutrophils by inducing apoptosis [62] and therefore to prevent subsequent tissue damage [63]. We are able to confirm significant activation of uterocalin expression in mammary involution (Fig. 6b) and the concomitant absence of neutrophil antigens (Table 3). In conjunction with the induction of uterocalin, clusterin (Sgp2), oncostatin M receptor (Osmr) and secretory leukocyte inhibitory protein (Slpi) were also increased (Fig. 6b). Clusterin protects cells against oxidative and chemical damage [33, 34] and has been suggested to be coexpressed with uterocalin as part of a localized acute-phase response in order to limit the damaging effects of residual neutrophils in involuting mammary gland [29, 63]. SLPi, an antichymotrypsin, also contributes to the elimination of neutrophil function, whereas Osmr may possess similar anti-inflammatory effects through the sequestration of proinflammatory cytokines. Osmr, Slpi and Sgp2 are all directly induced by Stat3 in mammary epithelial cells (Boland M, unpublished data) [64], whereas uterocalin expression conformed with Stat3 activity (Fig. 3 and 6b). A putative anti-inflammatory role for Stat3 was suggested in our previous study of Stat3-deficient mammary glands [17], in which we observed an increase in mastitis in the absence of Stat3. A similar protective mechanism may exist during parturition, when the LIF–Stat3 signal is also active [24, 65]. We are currently investigating the possibility that LIF–Stat3 signalling is responsible for the induction of these four genes in mammary gland, therefore confirming a dual role for LIF in involuting mammary gland. In addition to the direct effects of these anti-inflammatory acute phase genes, immunosuppressive factors including transforming growth factor-β (Inv3; Table 3) expressed by phagocytic macrophages are likely also to contribute to the suppression of an inflammatory response [64, 66].

Soluble defence factors and cells that are involved in humoral immunity are thought to be induced to prevent mastitis during this extended period of milk stasis. For example, the iron chelator lactoferrin is induced in mammary involution (Fig. 6c), which plays a role in mammary defence by protecting against coliform infection [36]. Indeed, low levels of lactoferrin is a predisposing factor for mastitis in humans [38]. Immunoglobulins not only have an important role in supplementing milk but are also required to facilitate opsonization by macrophages and neutrophils (IgM and IgG₂) [67] and the direct sequestration of bacteria (IgA).

Cd14 exhibited a dramatic pattern of expression throughout the complete time course (Fig. 6c). Cd14 is a core component of innate immunity, which binds LPS and augments LPS responses [42]. It is found bound to intramammary monocytes or neutrophils [41] in cattle subjected to endotoxin, and its soluble form is a known constituent of milk, in which it helps Cd14-negative cells respond to infection [68] and contributes to inhibition of gastrointestinal infections in human infants [69]. It may also play a role in recognition and phagocytosis of apoptotic cells by macrophages expressing cell surface Cd14 [70]. Thus, the role of elevated Cd14 may be both to prevent infection and potentiate apoptotic cell clearance. Binding of soluble Cd14 to LPS is augmented by the acute phase protein Lbp, which also exhibited an I_P expression profile (Fig. 6c).