“It is now well established that some micro-organisms can, under certain conditions, be deprived of all visible signs of life and yet these organisms are not dead, for, when their original conditions are restored, they can return to normal life and activity” ¹ .

“Bacterial populations in both batch and continuous culture are much more heterogeneous than is normally assumed, and such cultures may consist of several types of subpopulations simultaneously differing in viability, activity and integrity of the cells” ² .

Consider a typical axenic flask or broth culture of bacteria ( Figure 1), arguably the staple of modern laboratory microbiology. We seed a suitable growth medium with an appropriate inoculum of cells known to be capable of replicating in that growth medium. After a lag phase the number of culturable cells (the ‘viable count’ ^{3, 4} , as judged by plate counts of the number of colony-forming units observable on the same medium solidified by agar or a similar material) is observed to increase, typically exponentially, for a number of generations (the growth phase or exponential phase). Apart from the changes in nutrient concentration, and for non-synchronised cultures, it is generally taken that cells pass smoothly through their cell cycles en route to doubling their numbers by binary fission. The population distribution of organisms in different parts of their cell cycle during the exponential phase is thereby unchanged and thus in a steady state (from which the cell cycle parameters can even be inferred ⁵ ). In time this increase in cell numbers ceases, usually because of the exhaustion of a nutrient in a closed system, or sometimes in part or whole because of the build-up of toxins. Again, after a further period, the viable or colony count decreases (often to quite low levels if such starvation is carried out for extended periods). Inoculation of a new broth culture with a similar number of viable cells from this culture usually provides a simple repeat of the previous culture ⁶ , and in the absence of mutation may reasonably be anticipated, for organisms proliferating asexually, to be played out indefinitely.

The development of continuous ⁷ , nutrient-limited (‘chemostat’ ⁸ ) or feedback-controlled (‘turbidostat’ ^{9– 11} ) cultures was and is entirely consistent with this view of steady-state microbial doubling via homogeneous cell cycles that are common, within statistical fluctuations, to each cell. The same is true for cultures undergoing serial transfer (where there is slightly more of a focus on selection for genotypic variants that grow faster – see e.g. 12– 14).

There should be nothing controversial in the above passage, but in fact it hides a variety of assumptions that themselves conceal a considerable feast of very interesting physiology. The chief one here is that – given that all cells in the culture are genetically homogeneous and see the same ‘environment’, and modulo where they are in their cell cycles – all such cells are indeed supposed to represent a single population (as per Figure 2). If they do not, and as we shall see they never do ^{15– 18} , we are dealing with differentiated systems. It turns out that a particular subset of typical cell cultures – a phenotypically dormant or non-growing sub-population, occurring even in non-sporulating bacteria ² – is widespread to the point of ubiquity. This leads to an exceptionally important biology with significant consequences both for our understanding of microorganisms and our ability to harness and domesticate them. Although the relevant literatures rarely cite each other or overlap, it is clear that similar phenomena are common to bacterial behaviour in the natural environment, the laboratory, and in a variety of samples of clinical interest. This theory or hypothesis that we develop here comes about from the synthesis ¹⁹ of a large amount of data, and is summarised in Figure 3 and Figure 4.

While dormancy and resuscitation of rotifers had been observed by Leeuwenhoek himself in 1702 ¹ , some of the earliest modern indications for a physiologically significant ‘phenotypic heterogeneity’ ²⁰ or differentiation of microbial cultures came in the 1940s. In a conceptually simple experiment (illustrated in Figure 5), Bigger ²¹ exposed staphylococcal cultures to concentrations of penicillin that would normally be sufficient to kill them completely (and they did kill all but 1 in a million). However, these (10 ^-6) survivors, that Bigger ²¹ and McDermott ²² (and many modern commentators have) referred to as ‘persisters’, were not genetic mutations selected for resistance to penicillin, since when they were inoculated into fresh broth they were just as susceptible as were those in the first culture. Bigger recognised (correctly) that the only explanation that made any kind of sense was that despite being exposed to nominally the same conditions, these cells were operationally dormant in the sense of not replicating in a medium that, apart from the penicillin, would normally admit their growth (even if they were metabolically active ^{23, 24} ) and thus phenotypically resistant to the penicillin (that anyway kills only dividing cells ^{25– 27} ). Similarly, Luria and Latarjet ²⁸ noted that approximately 1% of the cells in a culture of Escherichia coli displayed a phenotypic resistance to normally sterilising doses of ultraviolet irradiation. Many similar experiments since (e.g. 29– 32), discussed in more detail below, have recapitulated this basic phenomenon. (We note here that high-frequency antigenic ‘phase’ variation can occur due e.g. to changes in microsatellite DNA ³³ ; detailed discussions of such genotypic changes ³⁴ , including those that can affect the extent of dormancy in persistent bacteria ³⁵ , are outwith the scope of the present, purely phenotypic analyses.)

For the avoidance of doubt, and in accordance with Keilin’s description with which we opened, we shall define dormancy as:

“a reversible state of {often} low metabolic activity, in which cells can persist for extended periods without division; we shall see that this often corresponds to a state in which cells are not ‘alive’ in the sense of being able to form a colony when plated on a suitable solid medium, but one in which they are not ‘dead’ in that when conditions are more favourable they can revert to a state of ‘aliveness’ as so defined” ² .

We thus stress ³⁶ the recognition that dormancy is not solely an innate property of a bacterial cell; it is a property assessed by one or more experiments, so whether a cell appears to be dormant depends on both the cell and the experiment used to assess that dormancy. (This principle shares a similar philosophical foundation to the independence from any specific experiment, or otherwise, of the perceived state of objects within the quantum theory ^{36– 38} ). As do Postgate ^{3, 4, 39} and Barer ^{40– 44} , we take the hallmark of a viable or living bacterial cell to be its ability to replicate or its ‘culturability’. This means that we cannot tell via culturability that a cell is alive, only (after a cell division) that it was alive ^{36, 45} . Dormant cells – even if ‘not immediately culturable’ – must by definition be resuscitable to form culturable cells. We also recognise (as does Michael Barer ⁸⁸⁹ ) that it may be hard to discriminate the resuscitation of dormant cells from the recovery of injured cells. Although the term ‘nonculturable’ is quite commonly used to describe not-immediately-culturable cells it is best avoided, as we cannot try every possible combination ⁴⁶ of incubation conditions that might serve to resuscitate a cell in a sample. ‘Non-cultured’, ‘as-yet-uncultured’ or ‘operationally nonculturable’ are better terms. Culturable, (operationally) non-culturable and (operationally) dormant bacteria in the differentiated bacterial (cellular) system can therefore be seen as distinct subpopulations of the system, and culturable and dormant bacteria as reversible states of the same population. A culture containing several subpopulations, whether distinct (as in Figure 2, or part of a single population characterised by a particular value from a range of an extensive variable) may be said to be differentiated (and of course may de-differentiate) in terms of physiological macrostates, that may or may not be able to interconvert. However, we recognise (thanks to Michael Barer ⁸⁸⁹ ) that such interconversion does not imply a mechanistic reversibility. The same kinds of issues attach to cells described as having any other physiological property with regard to the ability to replicate. We note (with thanks again to Michael Barer ⁸⁸⁹ ) that it is easy to conflate dormancy and ‘persistence’, since they do share some similarities (e.g. such cells are not immediately replicable); however, there is not much in the way of evidence as to how different say their expression profiles are, since it would require, for instance, single cell omics measurements, that are only just becoming available (e.g. 47, 48), more typically ⁴⁹ for the much larger eukaryotic cells. Certainly there can be extensive changes in gross biochemical composition as cultures are starved ⁵⁰ . One strategy would be to separate sub-populations ^{51, 52} , acquire ‘averaged’ values of say their transcriptome, proteome or metabolome, and see how much they differed. In a similar vein, whether states such as dormancy are adaptive is a matter for experiment.

The general relationships between various subpopulations of the bacteria within a differentiated cellular system are shown in Figure 6.

It has long been known that the number of bacteria observable microscopically exceeds, typically 100-fold, those that can readily be grown axenically in standard isolation media (i.e. to proliferate in liquid culture or to form colonies on solid media). The latter has been referred to as ‘the great plate count anomaly’ ⁶⁶ , and has been amply confirmed by more modern, culture-independent sequencing methods. A selection of papers and reviews serve to document both the numerical anomaly and the much greater biodiversity detectable by sequencing (e.g. 67– 86). It is thus useful to discriminate (1) bacteria that have been cultured, that are typically available in culture collections, and whose growth requirements are known, from (2) bacteria that may be recognised as novel via macromolecular sequencing (typically of ribosomal DNA ^{80, 87– 90} ) but that have not yet been cultured and whose growth requirements may not yet even be known. Much (sequencing) evidence indicates that the bulk of the ‘missing microbes’ or ‘dark matter’ ^{91– 93} in natural ecosystems falls into this second category ⁹⁴ , and that ‘single cell’ methods may be required to culture them ⁹⁵ .

There are at least four general reasons of principle why these organisms have not yet been cultured. We consider each in turn (although more than one may contribute in individual cases).

A related and unexpected discovery came from analyses of starved laboratory cultures of the actinobacterium Micrococcus luteus, in which almost all cells lost culturability ^{2, 198– 200} . However, they were not dead but dormant, as they could be resuscitated by using a combination of weak nutrient media and a signalling molecule found in spent culture supernatants ^{201– 206} . The original studies used flow cytometry to discriminate the physiological state of individual cells ^{51, 207– 210} (see also 211, 212). By using another ‘single cell’ assay based on dilution to extinction (that avoids artefacts connected with the regrowth of ‘initially viable’ bacteria ³⁶ ), we were able to purify the signalling molecule. It turned out to be a protein, named Rpf (for ‘resuscitation-promoting factor’) ²¹³ . In M. luteus there is only one homologue ²¹⁴ , and the gene (product) is essential for both resuscitation and multiplication ^{213, 215} . Rpf contains a highly conserved 70 amino acid ‘Rpf domain’ and is widely (and probably ubiquitously) distributed throughout the actinobacteria ^{216– 219} , but with examples elsewhere ^{220, 221} . Most organisms that have a homologue have more than one. Thus M. tuberculosis has five homologues ^{222– 224} . Rpfs can have peptidoglycanase and muralytic activity ^{225– 230} and known crystal structures are consistent with this ^{231– 236} . These activities can certainly account for at least some ²³⁷ of the resuscitation-promoting properties. As an extracellular protein that may be required for growth, and with a high level of immunogenicity ²³⁸ , it is obviously an excellent candidate target for inclusion in appropriate vaccines against pathogenic actinobacteria ^{213, 225, 239– 246} . It is also more directly of potential utility in stimulating bacterial communication and resuscitation in a variety of cultures in both samples taken from Nature ^{247– 257} and in the laboratory ^{258– 271} .

Having established the frequency of occurrence of microbial dormancy in the natural environment, it is of interest to understand better the mechanisms by which microbes might effect this dormancy and potential resuscitation. Unsurprisingly, microbiologists have turned to E. coli, and considerable progress has been made ^{24, 272– 279} .

The starting position is as in Figure 1 and Figure 6, to the effect that at any given moment in a typical culture a small fraction of the population is non-growing, and thus potentially dormant. Since clearly the same fraction cannot (or is wise not to) remain in dormancy indefinitely in the presence of suitable nutrients that permit the growth of its siblings, we must invoke at least one mechanism that can cause the bacteria to ‘oscillate’ between growing and dormant states. Many simple gene expression network topologies admit this behaviour ^{159, 280– 284} , including a simple feedback loop with delay ^{285, 286} , and we note that even whole cultures can exhibit oscillations and deterministic chaos ²⁸⁷ . While flow cytometric observations (e.g. 51, 288) show that even ‘homogeneous’ laboratory cultures show highly heterogeneous distributions in cellular volume (not just between X and 2X) and expression profiles (and see 289), our particular focus will be on ‘binary’ or ‘bistable’ systems in which individual cells either are or are not operationally culturable.

Experimentally, it is also common to assess the phenotypic ability of subpopulations of cells to tolerate normally inhibitory concentrations of bactericidal drugs ^{290, 291} , this being a marker for that fraction of cells that is ‘persistent’ (and maybe dormant) at the stage in question. Note that the persistence phenotype is not induced by the drugs ²⁷⁵ . Changes or transitions in the state of a particular cell in a population between the various phenotypic states is a phenomenon that may be (and is commonly) referred to as ‘phenotypic switching’.

Until relatively recently, almost all of clinical microbiology ^{331, 332} was based on rather classical methods of plate counting ³³³ , coupled to assessment of antibiotic sensitivity. Various means of automated blood culture that assess metabolism exist (although they require typically 48–72h to show a ‘positive’) ³³⁴ . Positive tests, often implicitly involving culture (and not just metabolism) within the assay, would be followed by other tests seeking to identify the organisms detected, nowadays typically by nucleic acid sequence-based methods ^{58, 335– 338} . However, these and other tests for the presence of antigens or even antibodies ³³⁹ cannot speak to the question of culturability (and of course antigens such as lipopolysaccharide (LPS) are shed by dying cells). This said, it makes little sense to try to culture microbes from samples that molecular sequencing methods indicate lack them, so the molecular methods always provide a useful starting point for seeking to resuscitate any resuscitable (hence operationally dormant) microbes that might be present.

The existence of bacterial DNA in even ‘healthy’ blood has long been known ³⁴⁰ , and since naked DNA would be degraded and living cells would soon kill the host, the (seemingly) obvious conclusion that the prokaryotic DNA must reflect occult, and potentially dormant, cells seems neither to have been drawn nor acted upon.

The idea that (typically intracellular) dormancy is a major component in some infectious diseases (including in the absence of antibiotics that may serve to light up ‘persisters’) is of course well-established, and the main purpose of this brief section is simply to remind readers of this. Such a reminder serves as a prelude to a longer discussion of the very many clinical circumstances where we consider that the role of dormant microbes is not widely appreciated, and where they are not really considered to involve a communicable or microbial component at all. Thus Table 1 shows a few organisms (and references) for which we consider that most readers would regard the idea of and evidence for dormancy as more or less uncontroversial. We do not include disease-causing infectious agents where they are better known for their ability to persist in the natural environment. Organisms such as Legionella pneumophila that represent significant public health issues, fall into this category, and Legionella and other persisters (in environments such as water system biofilms) are indeed well known (e.g. 341– 345), although they too have special adaptations to an intracellular lifestyle (e.g. 346).

As noted above for environmental microbiology, dormant bacteria can represent as much as 99% of the organisms that may be observed microscopically or by macromolecular sequencing, but classically (and by definition) they are not enumerated by culture-based methods that determine ‘immediate culturability’ ³⁶ . Such culture-based methods are also widely used in clinical microbiology. However, if we were to plate out 100 μL of a culture containing 200 bacteria.mL ^-1, of which 99% were dormant at any instant, we would expect (based on a Poisson distribution) to see fewer than 1 propagule or colony-forming unit per sample. We have noted above that it can be determined by sequencing that many of the non-cultured environmental organisms largely differ from those in standard culture collections. Certainly the examples given above in clinical microbiology, such as Tropheryma whipplei, were both observed microscopically and were sequenced prior to being brought into axenic culture.

The PCR method is exquisitely sensitive (down to one cell or propagule per sample), and we note that contamination artefacts from the PCR reagents represent a real issue that must always be checked (e.g. 375– 379), albeit this is no less true of blood cultures ³⁸⁰ . We have rehearsed elsewhere ⁶² five classes of argument that collectively make it implausible that these are all contamination artefacts; probably the most persuasive is simply the sheer number of prokaryotic DNA molecules that can be measured in blood and serum (e.g. 381– 383). While some of the most recent nucleic acid sequencing methods (e.g. 384– 389) do operate on single molecules, and genome-wide sequencing may soon be routine (e.g. 390, 391), the analysis of prokaryotes usually used a broad-range PCR step to amplify small-subunit rDNA to assess their presence, whether in environmental ^{74, 80, 88, 392} or clinical ^{388, 393– 405} samples. Using this, and while these methods alone cannot tell whether they were operationally dormant or dead, a very considerable number of studies have been performed in which ‘culture-negative’ clinical samples showed the presence of prokaryotes (at least as judged by sequence-based methods). This has some profound consequences.

We note that in a steady state such cells must be supplied at a rate equal to that of their clearance, and that the fact that clearance is lower than probably expected implies a significant ability of such cells to evade the innate and adaptive immune systems. We also take it that at least for common organisms (not very slow growers such as certain mycobacteria) the former rates must be much lower than those typically attainable in laboratory cultures, else we would have classical sepsis, and we do not. Most likely the observable facts are best accounted for by a combination of a periodic resupply of resuscitating cells, coupled to physiological changes in non-growing cells (especially including of cell wall antigens) that help them evade natural clearance mechanisms.

While PCR-based methods have long been used to assess the species involved in culture-positive samples ⁴⁰⁶ , e.g. from blood, our interest here is in samples that are culture-negative ⁴⁰⁷ that may yet (and indeed likely do) contain dormant cells. Among the first such indications of this was the study by Relman’s group ³⁴⁰ , who showed that the blood of even healthy controls contained significant amounts of prokaryotic DNA. Table 2 lists some studies in which broad-range PCR has been used to amplify and detect prokaryotic rDNA in culture-negative samples.

In environmental microbiology, as mentioned above, there were many early indications (as observed microscopically or flow cytometrically) for the presence of bacteria that did not (or not easily) prove resuscitable or culturable. In a similar vein, many studies have shown microscopically observable organisms in culture-negative but disease-positive samples. This is true both for diseases considered to be due to microbial pathogens and, in fact, for many others normally considered non-communicable ⁶² .

Microscopic observations in tissues have been a major part of the discovery process by which certain bacteria were indeed identified as the cause of various diseases. Billings ⁴³¹ , Price ⁴³² , Domingue ^{413, 433– 435} , Mattman ⁴³⁶ , Ewald ⁴³⁷ and Onwuamaegbu and colleagues ⁴³⁸ review the extensive and largely forgotten early literature. Domingue and Schlegel ⁴³⁹ also mentioned that they could recover culturable bacteria, probably mainly from L forms (see 62, 436, 440), from lysates of normal and diseased blood. It was to be assumed that these cells were not replicating at significant rates in the blood itself. However, we can find no evidence that this was ever followed up. Our own work ^{441, 442} , summarised in 62, showed that both bacillary and coccoid cells could be found attached to and within the erythrocytes of patients with Parkinson’s disease and Alzheimer’s disease, at rather greater concentrations than in samples taken from nominally healthy controls.

In a similar way, our preliminary data show that bacteria are visible in plasma, as well as in whole blood smears in various inflammatory conditions. Here we show bacteria in platelet-rich plasma (PRP) taken from a patient with systemic lupus erythematosus and smeared onto a glass cover slip ( Figure 7A and Figure 7B). We also show the same from patients with hereditary hemochromatosis ( Figure 7C) and type 2 diabetes ( Figure 7D). We also noted microbiota associated with erythrocytes in thromboembolic ischemic stroke ( Figure 8A and Figure 8B). (Our microscopy methods are as published previously (e.g. 442– 451), but fuller publications will appear elsewhere). The ultramicroscopic evidence that these are indeed small bacteria and not say, cellular debris or microparticles (see 452) is presently mainly morphological, though we note the considerable evidence for the presence of bacterial DNA in blood (see previous sections and e.g. 63, 340, 453).

It is worth rehearsing the very great significance of this. With erythrocytes being present at some 5x10 ⁹.mL ^-1 in human blood, even if only one erythrocyte in a thousand harboured just a single dormant bacterium (that would be hard to detect microscopically, but see 453– 457), the dormant bacterial load would still be 5,10 ⁶.mL ^-1. This is both far from negligible, and serves to exclude the (always potentially worrisome) claim that ‘it is all contaminants’.

A recent and highly significant paper by Damgaard and colleagues ⁴⁵⁸ bears discussion. These workers note ⁴⁵⁸ that while bacterial growth can normally be elicited during sterility testing in vitro from fewer than 1 in a 1000 blood units ^{459– 461} , transfusion-transmitted infections occur with a very much higher frequency (more like 10–12% ^{462, 463} , or even more ⁴⁶⁴ ), and are responsible for a high fraction of transfusion-associated deaths ^{465– 467} . Although it was acknowledged that venepuncture-associated contamination or an effect of transfusion in suppressing the immune system might contribute, it was also recognised ⁴⁵⁸ that one means by which to account for this would be that ‘normal blood’, and in particular its erythrocyte components, might also contain infectious agents that might be able to grow post-transfusion. Indeed, these authors found ⁴⁵⁸ that under anaerobic conditions a small number of colony-forming units (ca 4–5.mL ^-1) could be recovered by direct plating from fully 62% of blood units, with ‘controls’ producing an average of just 1 cfu.mL ^-1. More of the bacteria were associated with red blood cells than with plasma, and rDNA was used to identify them. These data are entirely consistent with the idea that dormant bacteria are present in the blood of even ‘normal’ individuals (note that periodontitis was not a criterion for donor exclusion here ⁴⁵⁸ ), that they are probably lurking in or on erythrocytes ^{468, 469} , and that they can be resuscitated and grow under the correct conditions.

We have surveyed the literature for evidence in which a microbial component has indeed been observed to be an accompaniment of, and probably a major contributory factor to, a variety of (typically inflammatory) diseases that are normally considered ‘non-communicable’. Rarely has the physiological state of these microbes been considered, but since it would be obvious if they were growing, it is most likely that they are indeed dormant. Table 3 summarises these highly extensive associations. While some are just associations, and we could have extended this table considerably, some studies (e.g. 470) contain very detailed aetiological arguments that leave little room for doubt. Overall, the sheer size of the Table does strongly indicate the commonality of many of the microbially based mechanisms underpinning or accompanying various autoimmune and inflammatory diseases. In conditions such as atherosclerosis, transient ischemic attacks (TIAs), and stroke, it is very easy to conceive how resuscitating bacteria might serve to block the flow of blood, for instance. At all events, our main point here is that the evidence for a microbial contribution to many diseases supposedly lacking a microbial component is both multi-factorial and very considerable. Indeed, the purpose of a synthetic review such as this is to provide such pointers for more detailed studies in individual cases. Our specific interest is with the chief mechanisms by which these supposedly dormant bacteria might resuscitate and act as triggers of disease.

Many of the diseases in Table 3 are precisely those inflammatory diseases that we have listed before as coupled to iron dysregulation ^{183, 184, 449, 452, 733} . A consequence of our analysis is that iron dysregulation and sepsis (as judged either by genuine infection by culturable bacteria or their inflammatory products such as LPS) should be associated causally with these various other diseases.

This leads to a variety of predictions and postdictions that we rehearse. A purposely simple (and simplistic) indication of a plausible chain of events (for which each step is underpinned by substantial evidence) is given in Figure 9, both in general terms (for unspecified diseases) and for a couple of steps to type 2 diabetes. Figure 9 aims specifically to highlight the relationship between the ability of available iron to stimulate bacterial growth and the potential disease sequelae thereof.

We have here brought together some of the relevant elements of environmental, laboratory, and clinical microbiology. We have argued that while their languages may differ (e.g. ‘dormancy’ vs ‘persistence’), very similar phenomena have been observed in each of these spheres (plausibly underlying a commonality of mechanism). Certainly the ability to culture microbes, and not merely to observe them (whether microscopically or via their macromolecular sequences or chemical products), remains an important goal of basic microbiology. This is likely to have significant payoffs in bioprospecting (e.g. 179, 806). However, we are sure that improved methods of detecting and identifying these dormant bacteria, whether this is done via chemical imaging, through macromolecular amplification and/or sequencing, or through resuscitation and culturing, will have a major role to play in increasing the awareness of their existence and importance.

Clearly dormant and/or persistent bacteria are likely to be relatively avirulent when they are in such dormant states, and able to bypass the attentions of the innate immune system (albeit the production of superantigens by at least some microorganisms ^{807, 808} may be what triggers autoimmune diseases). This ‘stealth’ antigenicity is probably why they have been largely unnoticed by us too ⁸⁰⁹ , and their routine estimation via molecular methods ⁸¹⁰ seems highly desirable. Indeed, virulence varies widely between individual strains (e.g. 811, 812). Modern molecular microbiology places much emphasis on the virulence of the pathogen, with concepts such as ‘pathogenicity islands’ ^{813– 818} , ‘virulence genes’ ^{819, 820} , and the ‘virulome’ ⁸²¹ being commonplace. However, if dormant microbes resuscitate (or are to be resuscitated) in vivo we shall need to pay much more attention to the environmental triggers that can cause this to happen than we probably have so far ⁸²² (given that the pathogen genotype is fixed ^{823, 824} ). In other words, virulence, like dormancy, is a phenotypic as well as a genotypic property. We remain largely ignorant of the means by which an optimal immune system has been selected for (or against) by longer-term evolution on the basis of microbial exposures in early life, and how this may have changed with more recent changes in human lifestyle ^{825– 828} . Nor do we understand how such microbes might enter and exit blood cells (and see 62, 347, 829– 833) (albeit the known endosymbiotic origins ^{834, 835} of eukaryotic organelles must have presaged such mechanisms). Similarly, we do not yet know what may cause these dormant microbes to resuscitate (and/or to exit their intracellular niches). However, the potential for iron-associated replication and (e.g.) LPS production and shedding does provide a very straightforward explanation for the continuing low- or medium-grade inflammation characteristic of the many inflammatory diseases we have considered here and elsewhere ^{183, 184, 449, 452, 733, 890} ( Figure 9).

Recognising that correlation does not at all equate to causality (e.g. 195, 836), one approach to Science is based on varying independently something considered a cause and observing its predicted effects (e.g. 195, 837, 838). Temporal covariation of measurands can also be performed. The levels of free iron are clearly one such possibility. To assess causality in microbiology it is usual (e.g. 815, 839– 841) to invoke what are (variously ⁸⁴² ) referred to the Henle-Koch or Koch’s postulates. These are based on the nature and presence, but not the physiological state, of an agent that might be believed to ‘cause’ (or at least contribute to) an infectious disease. Consequently, dormancy poses something of a challenge to the full completion of the required tests. Indeed a number of authors ^{437, 815, 842– 845} have recognised that these tests may need revision in the light of the ability to identify disease-causing microbes by sequence alone. We suspect that a key element here will be the ability to resuscitate dormant organisms in vivo and to see the effects of that on clinical disease.

From a ‘philosophy of science’ point of view (e.g. 841), one strategy taken to develop the evidence for a particular point of view hinges on the idea that if a series of ostensibly unrelated findings are brought together into a self-consistent narrative, that narrative is thereby strengthened. This is the strategy pursued here, and it is known as ‘coherence’ ^{846– 848} .

As phrased by Silvers ⁸⁴⁹ , “Several of our contributors showed how discoveries and insights could emerge with what seemed great promise, and yet be pushed aside, discarded, and forgotten – only to re-emerge once again, sometimes many years later, and become, in their new formulation, accepted as important”. In this sense, and as presaged in the opening quotation ¹ , it seems that ideas, as well as bacteria, can remain dormant for extended periods ^{850, 851} .