Alternative testing systems for evaluating noncarcinogenic, hematologic toxicity.

Hematopoietic tissues are the targets of numerous xenobiotics. Clinical hematotoxicity is either a decrease or an increase in peripheral blood cell counts in one or more cell lineages--a cytopenia or a cytosis, respectively--that carries a risk of an adverse clinical event. The purpose of in vitro hematotoxicology is the prediction of these adverse hematologic effects from the effects of the toxicants on human hematopoietic targets under controlled experimental conditions in the laboratory. Building on its important foundations in experimental hematology and the wealth of hematotoxicology data found in experimental oncology, this field of alternative toxicology has developed rapidly during the past decade. Although the colony-forming unit-granulocyte/monocyte neutrophil progenitor is most frequently evaluated, other defined progenitors and stem cells as well as cell types found in the marrow stroma can be evaluated in vitro. End points have been proposed for predicting toxicant exposure levels at the maximum tolerated dose and the no observable adverse effect level for the neutrophil lineage, and several clinical prediction models for neutropenia have developed to the point that they are ready for prospective evaluation and validation in both preclinical species and humans. Known predictive end points are the key to successful comparisons across species or across chemical structures when in vitro dose-response curves are nonparallel. Analytical chemistry support is critical for accurate interpretation of in vitro data and for relating the in vitro pharmacodynamics to the in vivo pharmacokinetics. In contrast to acute neutropenia, anemia and acute thrombocytopenia, as well as adverse effects from chronic toxicant exposure, are much more difficult to predict from in vitro data. Pharmacologic principles critical for clinical predictions from in vitro data very likely will apply to toxicities to other proliferative tissues, such as mucositis.

The field of hematotoxicology includes the study of adverse effects of toxicants on mature blood cells and also the precursor cells in the hematopoietic (blood forming) tissues. There are established techniques for assessing adverse effects of xenobiotics on mature blood cells. More recently, the availability of recombinant hematopoietic growth factors makes possible the evaluation of adverse effects against the bloodforming precursor cells as well. Thus, it is now possible to study human hematotoxicology in the preclinical setting. Because it is possible to study the effects of a toxicant on its actual target cell, it seems reasonable to expect in vitro hematotoxicology to be highly predictive.
Fundamentally, toxicology has two goals: identification of the tissues that are susceptible to the toxic effects of the xenobiotic and determination of the level of acute and chronic exposures (doses) that these tissues can tolerate without clinical consequences. The first goal is qualitative; it necessarily involves comparative toxicology in multiple tissues. This comparison is most efficiently completed in vivo where all organs can be exposed simultaneously. A similar evaluation performed ex vivo necessarily requires the assay of multiple human tissues under identical conditions, which at the present time is not technically possible. The estimation of the acceptable level of human exposure from in vitro hematotoxicology data assumes that the toxicant's most potent effects are toward the bone marrow, i.e., hematopoietic tissue is the most sensitive of the human tissues to toxicity. By its nature, hematotoxicology complements, refines, and actually improves standard toxicology testing (usually in vivo), which still is required to identify hematopoiesis as the most likely tissue target of the toxicant in humans. The second goal is quantitative and estimates the level of toxicant exposure that can be tolerated by the target tissue. This is critical for accurate risk assessment and establishment of reasonable regulatory limits on exposure. In this case, ex vivo evaluation must involve evaluation of toxicity to only the target tissue or cells derived from it and in vitro hematotoxicology meets this second goal.
The laboratory techniques for evaluating the effects of a toxicant on human hematopoietic tissue are relatively straightforward, and there is a wealth of data on the toxic effects of xenobiotics on hematopoietic target cells. Most laboratories have the capability of performing all of the ex vivo tests described here. In contrast, the interpretation of the data and the quantitative prediction of the acceptable level of clinical exposure from the data are much more difficult. It is this latter arena in which we have specialized. We have investigated prediction models that translate the data into predicted hematologic perturbations as a function of toxicant exposure levels. The goal in the regulatory setting usually emphasizes the prediction of two levels of exposure: the highest dose that will not cause a dinically adverse effect and the dose that causes maximally tolerated, reversible perturbations in peripheral blood counts. The former is often termed the no observable adverse effect level (NOAEL), whereas the latter is termed the maximum tolerated dose (MTD). Human exposure at the MTD is undesirable except for anticancer agents; with this exception, permissible exposure limits (PELs) are set for all regulated products from the NOAEL. Substantial pharmacologic issues arise during the prediction of clinical outcomes from in vitro data. Ex vivo assays determine the concentration and exposure duration that cause toxicity. In contrast, the regulation of human exposure requires that decisions be based on units of dose or dose intensity (i.e., mg/m2 or mg/kg per unit of time). The prediction of a dose that produces the in vitro concentration-time-effect relationship in vivo involves pharmacokinetics-the disposition of xenobiotic throughout the body. By analogy to animal and clinical toxicology, in vitro hematotoxicology aims to determine the no observable adverse effect concentration (NOAEC) and the maximum tolerated concentration (MTC) of toxicant under a specific schedule of exposure. The pharmacokineticist can then calculate the NOAEL and MTD doses that produce these NOAECs and MTCs. The NOAEC and the MTC are two end points useful in regulatory science sought during in vitro hematotoxicology studies.
The success to date lies primarily in the identification of the in vitro inhibition concentration value that is the MTC. An international validation study sponsored by the European Centre for the Validation of Alternative Methods (ECVAM) is underway to evaluate the predictive value of this putative MTC for clinical neutropenia (1). Results from this formal evaluation of in vitro hematotoxicology can be expected by January 1999. Because a wealth of detailed preclinical and clinical pharmacokinetic and hematotoxicity data exists for antineoplastic agents, these compounds serve as prototype hematotoxicants to elucidate the principles of predicting the NOAEC and the MTC. Once the prediction models are developed with antineoplastics, it is expected that they will generally be applicable to all other xenobiotics.
The Hematopoietic System as a Prototype for in Vivo and in Vitro Toxicology Circulating Blood Cells and the in Vivo End Point The blood of mammals contains a variety of differentiated cell types with specific functions. Red blood cells (erythrocytes) deliver oxygen; platelets contribute to clot formation; and lymphocytes, monocytes, and granulocytes provide resistance to infectious organisms and foreign materials. The granulocytes and platelets have a halflife in the circulation of only a few hours and a few days, respectively, whereas erythrocytes have a much longer half-life of several months. The proportion of white blood cells that are lymphocytes or granulocytes differs across mammalian species, and this must be kept in mind when interpreting toxicologic data. For example, rodents exhibit a lymphocyte:granulocyte ratio of about 4:1 whereas humans and dogs show a 1:4 ratio. A toxicant that causes severe neutropenia in the absence of lymphocytopenia will only be detected during rodent toxicology studies if differential counts are made on the leukocyte population. Otherwise, the maximum adverse effect will be a 20% reduction in white blood cell counts, which is likely to be considered unremarkable. Furthermore, it is important that validation studies of laboratory end points correlate the in vitro data to the correct in vivo end point. Assays to predict neutropenia need to be correlated with neutrophil counts, not with leukocyte counts. Hematopoietic Cells as Producers ofNew Blood Cells New blood cells must constantly be produced to maintain the peripheral blood cell counts, and under healthy physiologic conditions there is a balance between new cell production and cell loss that leads to the constancy of blood cell counts observed clinically. The hematopoietic tissues that produce new blood cells are primarily located in the bone marrow in humans but in both bone marrow and spleen in rodents and sometimes dogs. Hematopoietic tissue contains a continuum of increasingly differentiated elements within all blood cell lineages. Developing myeloid, erythroid, megakaryocytic, and lymphoid cells can be distinguished cytologically by the trained eye or with histochemistry for lineagerestricted features. The highly undifferentiated cells are progenitors of the immature blood cells. There are also hematopoietic stem cells in this undifferentiated population. Fixed stromal cells (both fibroblastic and histiocytic) and T lymphocytes in these tissues may exert some regulation over hematopoiesis.
In addition to these lineage-restricted progenitors, hematopoietic tissue contains immature progenitors that form colonies containing multiple myeloid lineages (3642). One such progenitor is CFU-granulocyte, erythroid, megakaryocyte, and monocyte (CFU-GEMM). As the name implies, colonies formed by this pluripotent progenitor are distinguished by the presence ofcellular elements from all of these lineages.
Several CFUs lie close to the hematopoietic stem cell developmentally. The high proliferative potential-colony-forming cell (HPP-CFC) produces a very large colony that contains primarily cells with a blastlike cytology, a small proportion of which will reform a clonal colony after Environmental Health Perspectives * Vol 106, Supplement 2 * April 1998 recloning (43)(44)(45)(46)(47)(48)(49)(50). The long-term culture-initiating cell (LTC-IC) is a cell with multilineage potential found at low frequency in the marrow, probably positioned prior to the HPP-CFC in myelopoiesis (51)(52)(53)(54). LTC-IC cells exhibit several characteristics of stem cells, including some capability of self-renewal, maintenance of both lymphopoiesis and myelopoiesis, and long-term reconstitution of a lethally irradiated host (55)(56)(57)(58)(59)(60)(61)(62)(63)(64). Bone Marrow Stroma In contrast to the proliferation and expansion of tissue mass by hematopoietic cells, the primary function of the stroma is the nurture and support of developing blood cells (52,(65)(66)(67). However, there is a stromal colony-forming unit called CFUfibroblastoid (CFU-f), which produces a colony composed of adherent cells exhibiting morphologic features of fibroblasts, adipocytes, and other stromal elements (65,68). Although proliferating CFU-f is highly sensitive to many bone marrow toxicants in vitro, it is unclear whether the CFU-f is an in vivo target cell of toxicants that affect replicating cell types.

Interactive Cultures of Hematopoietic and Stromal Cells
Cultures of hematopoietic cells combined with stromal support cells can be maintained for several months in the absence of exogenously added cytokines by adhering to prescribed methodology (69)(70)(71)(72). These so-called Dexter cultures sustain myelopoiesis at nearly steady-state levels for several weeks, after which time myeloid cell output begins to dedine. It is possible to quantitate the output of progenitors or mature myeloid cells over time in these cultures. Likewise, longterm marrow cultures called Whidock-Witte cultures sustain lymphopoiesis (73)(74)(75)(76).
Modifications of cell culture conditions can switch the cultures between lymphoand myelopoiesis, making it possible to evaluate both lymphoand myelopoiesis from the same cell culture (56,77). Because cytokines are not added to these cultures, they are thought to model closely the steady-state hematopoiesis that occurs in vivo. However, the progressive dedine in mature cell output indicates the need to develop culture methodology for maintaining hematopoietic tissues in a homeostatic state.
Toxicity to the Mature Blood Cell Compartment Toxicity affecting primarily the mature, differentiated compartment usually manifests clinically as rapid-onset cytopenia. An example of this toxicity is phenylhydrazineinduced anemia caused by direct hemolysis of erythrocytes (10). However, hematologic toxicity following xenobiotic exposure can also manifest clinically as leukocytosis. Eosinophilia associated with exposure to IL-2, lithium, or a contaminant in overthe-counter preparations of L-tryptophan is due to toxicant-induced secretion of hematopoietic growth factors from mature blood cells (94)(95)(96)(97)(98)(99)(100)(101). Xenobiotic-induced leukocytosis can also be due to abnormally high neutrophil or lymphocyte counts (102)(103)(104)(105). CFU assays usually do not play a critical role in investigating this type of xenobiotic toxicity unless it cannot be explained by increased cytokine secretion.
Toxicity to the Progenitor

Compartment
The mechanism of hematotoxicity most frequently and thoroughly studied in vitro is the acute effects of toxicant on marrow progenitors like CFU-GM (referred to as CFC-c in the older literature) and CFU-Mk. Toxicity is quantified from the number of surviving progenitors as a function of exposure level under maximally stimulatory cytokine concentrations.
Drug exposure levels that inhibit colony formation by approximately 50% do not cause neutropenia clinically (129,148). It is likely that the in vitro and in vivo data fail to correlate at these mildly toxic levels of exposure because the hyperplastic response of the marrow can compensate for this magnitude of progenitor loss. For example, there may not be any hematologic consequence from a 2-fold reduction in the frequency of CFU-GM if balanced by a 2-fold increase in marrow cellularity (no net gain in total CFU-GM). A direct correlation may exist between the decreases in donogenic survival and absolute neutrophil counts only when toxicant levels cause such a severe lesion in the progenitor population that marrow hyperplasia cannot compensate.
In vitro assays can also be used to investigate cytopenias caused by noncytotoxic xenobiotics. For example, substances that reduce the number of differentiated blood cells produced per progenitor would be expected to decrease the size of the donal colonies (the number of cells per colony), but not the number of colonies. Colonies that resemble CFU-GM colonies but are too small are named clusters. Thus, reduced CFU-GM colony formation accompanied by a shift in the distribution of colony size toward dusters argues against destruction of the progenitor. Trichothecene mycotoxins appear to shift the balance between proliferation and differentiation toward the latter (113)(114)(115). Interferons, the transforming growth factor beta (TBF-P) family, and tumor necrosis factors inhibit colony formation by CFU-GM, CFU-GEMM, and HPP-CFC (149)(150)(151)(152)(153)(154)(155)(156). Experiments with TGF-Ps indicate how subtle yet specific these effects on progenitors can be (149)(150)(151)(154)(155)(156). The principles of predicting the parameters of neutropenia from in vitro data have not yet been established for this category of toxicants.
There is considerable confusion regarding a quantitative relationship between progenitor toxicity and acute anemia or thrombocytopenia. Megakaryocytopoiesis involves not only mitosis but also endoreplication, and it is unclear whether cell-cycle dependent toxicants are also endoreplication toxicants. For CFU-Mk assays, spontaneous colony formation occurs in vitro without Environmental Health Perspectives . Vol 106, Supplement 2 * April 1998 exogenously added cytokines; and colony formation is quantified based on a A value, i.e., the net increase in colony number due to cytokine stimulation. A cytokine-free control is included to quantify the number of colonies that form independently of added cytokine.
Toxicant-induced anemia is also difficult to predict from erythroid progenitor assays. Because erythrocytes circulate with a relatively long half-life, an acute toxicantinduced disruption of erythropoiesis is probably insufficient to cause anemia. For example, dipyrone inhibited not only CFU-GEMM and CFU-GM but also BFU-E in the presence of serum from a patient with drug-induced agranulocytosis and thrombocytopenia but not anemia (157). Also, ceftazidime is an equally potent inhibitor of BFU-E and CFU-GM, although it is clinically associated with agranulocytosis (158).
The most difficult hematotoxicity to predict with in vitro assays is the progressive loss of one or more blood cell lineages during chronic exposure to a toxicant: agranulocytosis, pancytopenia, and aplastic anemia. In some cases the toxicity leaves a permanent dysfunction, while in other cases the toxicity resolves after identification and removal of the toxicant. The distinction between irreversible aplasias like aplastic anemia and the progressive yet usually reversible aplasias like agranulocytosis is important for proper application of in vitro assays. In practice, toxicity may not be definable as reversible or irreversible if the toxic insult cannot be identified or if the degree of permanent damage is insufficient to cause permanent symptoms. In the cases of irreversible marrow damage after multiple toxicant exposures, it is often impossible to know whether toxicity would have occurred after a single exposure.
Progressive yet reversible xenobioticinduced cytopenia generally indicates a direct effect of toxicant or metabolite on hematopoietic cell types. CFU-GM levels in patients with drug-induced agranulocytosis are depressed relative to controls (91,(159)(160)(161)(162)(163)(164), and inhibition of myeloid and erythroid progenitors is a likely mechanism for beta-lactam-induced agranulocytosis (158) and contributes to phenylhydrazine-induced anemia (10). In these cases the in vitro progenitor assays correctly identified the mechanism of myelosuppression. Hypersensitivity of progenitors from susceptible individuals to the toxicant or a toxic metabolite can be demonstrated with progenitor assays (161,(165)(166)(167)(168)(169)(170)(171). However, other xenobiotics (chlorpropamide, phenytoin, methimazole, valproate, disopyramide, and phenacetin) not only inhibit progenitors directly but also induce both cellular and soluble immune mechanisms, which inhibit hematopoiesis (161,163,(172)(173)(174)(175)(176). Because immune constituents can contribute to hematotoxicity, it is important to eliminate effector cells when trying to prove a direct toxicity of xenobiotic to progenitors in vitro (168). Otherwise, the effects of inhibitory cytokines released by mature T cells and monocytes in response to toxicant may be measured instead, which would be a toxicity to the mature blood cell compartment rather than the progenitor compartment. Unfortunately, in many cases when the immune system contributes to hematotoxicity, in vitro studies using marrow from normal donors are not informative and in fact are usually inconclusive (167,170,177).
Microculture techniques for progenitors are also available when xenobiotic is in short supply (180).

Toxicity to Bone Marow Stroma
The microenvironment is a target for toxicant injury (90), and assays of stromal function have been used to investigate the effects of toxicant exposure. Bone marrow stroma is more sensitive than hematopoietic progenitors to the toxicity of acute neutron irradiation but less sensitive to X rays, and hematopoietic stem cell and stromal cell repopulation, but not progenitor survival, are dependent on dose rate (109,112,124,(181)(182)(183)(184). Long-term bone marrow cultures based on Dexter's method (69) have been used to investigate the agranulocytosis associated with ceftazidime and other agents (82,158). In addition, the CFU-f is potentially a dose-limiting cell type for radiation and some quinones of benzo[a]pyrene (183)(184)(185). CFU-f is the most sensitive progenitor to AZT toxicity, and AZT causes perturbations in parameters of long-term bone marrow assays at clinically relevant concentrations (186). Both conventional and high-dose cancer chemotherapies cause permanent damage to the marrow microenvironment (111,148,(187)(188)(189)(190)(191)(192). However, clinical hematology parameters can recover and even normalize in spite of continuing in vitro measurements of impaired marrow function (148,182,(189)(190)(191)(192)(193)(194)(195), casting doubt on the power of these in vitro assays to predict clinical outcome after chronic exposure.

Conceptual
Framework for in Vitro Hematotoxicology Studies Alternative hematotoxicology has benefited tremendously from the progress in experimental hematology over the past 40 years. Recombinant cytokines are available, and consequently a number of hematopoietic cell types can be assayed. In vitro hematotoxicology differs significantly from other alternative toxicologies in that the actual target cell of the toxicant can be studied in the laboratory. In addition, human exposure-hematotoxicity relationships are available from clinical trials in oncology for quantitative correlation with in vitro data.
Finally, there appears to be little disagreement over the standardized conditions to use for quantitative assays and several laboratories in this field have reached similar conclusions about experimental variables via independent studies (1,81,84,85,178,179). What Should Be Predicted? Clinical hematotoxicity resulting from xenobiotic exposure can be described by four parameters (Figure 1): severity of the change in blood cell counts (cytopenia or cytosis), time from exposure to the most severe toxicity, duration of the nadir, and time required for recovery from the toxicity (degree of reversibility). These parameters completely describe the clinical course and nature of xenobiotic-induced cytosis and cytopenia (88). These toxicologic parameters have several important characteristics. They are not linked to any assumptions about molecular mechanism; they are clinical end points measured in patients and therefore they constitute the in vivo data that in vitro end points must predict, and they provide a quantifiable description of xenobiotic effects on human blood cell counts.
Because regulatory agencies must make decisions about human PELs even for compounds with unknown mechanisms of action, it is preferable to have mechanismnaive assays, i.e., assays predictive across broad mechanistic classes. Molecular mechanism is not trivialized by this approach; rather it is admitted that this information is not required to regulate the product or treat the intoxicated patient. For example, any compound that caused reversible thrombocytopenia of 10 days' duration will be treated similarly, whether the xenobiotic disrupted Parameter 4, time required for recovery 1- assume that the dose-limiting cell compartment in vivo is the cell type most sensitive to the toxicant in vitro.
It is possible that by identifying the dose-limiting progenitor, we may predict the time to nadir. For the neutrophil lineage, this hypothesis postulates that neu-.tropenia due to a lesion in an immature neutrophil progenitor population will take longer to appear than a lesion in a relative late-stage neutrophil progenitor pool. This has yet to be proven, however.
In Vitro Pharmacology The in vitro assays are relatively simple to _* -*> Parameter 3, duration of cytopena perform and a wealth of data can be generated in a short period. In general, the results of the test will be expressed as the change in the clinical consequences of toxicant exposure. However, the way this goal has been ransduction or destroyed progenitors. toxicity to all of the cell types in the lineage phrased implies a subtle distinction from w mechanism-independent toxicolbut only the in vitro end point from the the conventional purpose of toxicology generally useful and advantageous assay of the actual target cell population studies. Traditionally, the question is one of ecause it prevents assay proliferation will be the most likely predictor of clinical classification: Is this compound a hematome, in which a new assay must be cre-toxicity. Given data from assays of mature toxicant? However, it seems reasonable to r each mechanistic dass that must be blood cell lysis and dysfunction, progenitor assume that any toxicant will be a hematoed, and because it can contribute to survival and function, and stem cell and toxicant if the other tissues of the body can ulation of new products with poorly stromal cell function, how does one predict tolerate high enough exposure levels.
what will be the clinical manifestation of Certainly there will be compounds that e goal of in vitro hematotoxicology is toxicant exposure?
inhibit CFU-GM in vitro but do not cause diction of these four parameters (in It is a well-established principle of in neutropenia, because exposure levels never Id points) from end points obtained vivo toxicology that the adverse effects of reach a high enough level: for example, the ssays of human hematopoietic funcexposure are determined by the most sensi-nonmyelosuppressive nucleoside analog ter in vitro exposure to the toxicant tive of the exposed tissues. The most sensi-dideoxycytidine is in fact toxic to CFU-Mk ). Different in vitro end points, and tive tissue is called the dose-limiting tissue, and CFU-GEMM in vitro (116,144,145). ps even different hematopoietic and in vitro toxicology must predict the Because toxicologic experience lends credimay be required to predict each in ramifications of exposure in this tissue to bility to this assumption, the practice of trameter. Most studies focus on the predict clinical toxicity. By analogy, the classifying chemicals as hematotoxicants :ion of the severity of the nadir.
manifestations of toxicant exposure in the should be abandoned.  (88). This analogy is helpful in the inter-the past 7 years. The PELs for most reguany hematologic lineage there are a pretation of in vitro data when a toxicant is lated products are based on the NOAEL r of potential target cells for the toxi-evaluated on the entire spectrum of cell dose, whereas antineoplastics are regulated he mature cell compartment in the types that constitute a hematopoietic lin-based on the MTD. We have come to tream, the progenitors in the marrow eage. For example, the nature of the neu-understand that the most useful in vitro end lake these blood cells, and the suptropenia caused by a toxicant will be points are those that predict the level of sysg cells in the hematopoietic tissues determined in large part by whether temic exposure to toxicants at these two gulate blood cell production. Most of mature neutrophils, immature neutrophils, doses of regulatory significance. Hence, the -an be assessed for toxicity in vitro myeloid progenitors, stem cells, or purpose of in vitro hematotoxicology is to is important to determine which is myelopoietic support cells are the in vivo predict the toxicant exposure levels associual target cell of the toxicant in vivo. target of the xenobiotic. Until more is ated with the MTD and the NOAEL doses. especially important because many understood from attempts at in vitro-in We have termed the predictive in vitro end its may show some degree of in vitro vivo correlations, it seems reasonable to points for these two exposure levels the Plag rather we should assume that they could, and determine the AUC at the NOAEL Shortexposuretime and MTD for the neutrophil lineage ( Figure 2B).
This understanding clarified how to determine if the CFU-GM assay could be validated for predicting neutropenia and Long exposure time used in the regulatory setting (88). We concentrated exclusively on this hematotoxicity 2 because the relationship between progenitor e.g. AUC numbers and peripheral blood cell counts is e.g., simpler for neutropenia than for anemia or thrombocytopenia. Furthermore, to identify the principles and concepts to include in dinical prediction models, we restricted our initial studies to the simplest of toxicity, the reversible neutropenia caused by an acute toxicant exposure. This rationale also focused the ECVAM validation study in hematotoxicology on the prediction of neutropenia (1).
sma AUC at MTD 1 2 Log toxicant exposure, e.g., AUC Figure 2. (A) Typical in vitro hematotoxicology data relating the decrease in a hematopoietic end point as a function of toxicant exposure in units of concentration, AUC, or whatever pharmacologic parameter of exposure is most closely linked to inhibition. The data shown here reflect the differential toxicity between a brief and long duration of exposure commonly observed with cell cycledependent toxicants. (B) In vitro toxicology attempts to relate a cell culture end point to a regulatory end point such as the NOAEL or the MTD. In vitro systems relate changes in end points to concentration-based measurements, and the estimation of dose relies on estimates of pharmacokinetic parameters. Several lines of evidence point to the AUC that inhibits CFU-GM by 90% (e.g., the AUC at IC90) as the predictive end point for the plasma AUC that will cause grade 4 neutropenia at the MTD ( Table 1). The in vitro end point that predicts the AUC at the NOAEL has not been determined.
MTC and NOAEC, respectively. This is analogous to the determination of the no effect and maximum tolerated doses in vivo, except that in vitro systems use concentration units instead.
For many hematotoxic compounds the severity of toxicity is related to a pharmacologic measure of exposure called the area under the plasma concentration versus time curve (AUC). Thus, the hypothesis of in vitro hematotoxicology is that the in vitro AUCs at the MTC and the NOAEC are identical to the plasma AUCs at the MTD and the NOAEL, respectively. This strategy shifts the emphasis of the in vitro study to end points that lie on the X axis, not the Y axis. We should not try to predict whether compounds inhibit neutrophil production:  (Table 1). A quantitative study of pyrazoloacridine found close correlation between inhibition of in vitro colony formation and the grade of neutropenia in vivo (129). In this study, the AUC that caused grade 3 to 4 neutropenia in humans inhibited in vitro colony formation of human CFU-GM by 90%. The result suggested that the 90% inhibition concentration (IC90) from the human CFU-GM assay was the MTC when this progenitor is dose limiting. This conclusion was confirmed and extended in the mouse by showing that the AUC at the IC90 from the murine CFU-GM assay was associated with a 90% reduction not only in absolute neutrophil count but also marrow CFU-GM (R Parchment, unpublished data). Subsequently, the IC90 was the in vitro end point that predicted the differential between human and mouse MTD for the topoisomerase I inhibitor topotecan, whereas other IC end points failed to predict the MTD differential from the nonparallel human and mouse concentration-toxicity curves. Preliminary results show that the IC90 ratio correlates to the MTD ratio for seven of seven tested drugs. Current studies are examining whether the IC90 is the MTC for other progenitor compartments and for antineoplastic agents with other mechanisms of action. It is possible that a 90% reduction in content might not be tolerated in other progenitor compartments, while confirmation of this putative MTC for CFU-GM and additional toxicants will help clarify prediction models and gain regulatory acceptance. Based in part on these studies, the ECVAM validation study of the CFU-GM assay will attempt to validate the IC90 as the MTC from which the plasma AUC at MTD can be predicted for xenobiotics associated with dose-limiting neutropenia (1).

PrdictingToxicant Exposure (AUC) at the NOAEL from in Vitro Data
Although the NOAEL is used much more frequently than the MTD as the basis for setting PELs for regulated products, there is a lack of understanding about the maximum degree of loss of progenitors that does not result in clinical neutropenia. This issue directly relates to our understanding of the resilience of hematopoietic tissue to toxic insult and the compensatory mechanisms, such as marrow hyperplasia, that can overcome mild reductions in progenitor survival. For example, a 50% reduction in CFU-GM survival coupled to a 2-fold increase in marrow cellularity would not result in any net loss of neutrophil progenitors. During our in vitro-in vivo correlation study in the Phase I trial of 9-methoxypyrazoloacridine (129), the greatest plasma AUC that did not cause neutropenia inhibited human CFU-GM by 35%. Although circumstantial, these results suggest that the IC35 may be the NOAEC, i.e., the end point from which the AUC at NOAEL can be predicted. Unfortunately, the minimum number of progenitors required to maintain normal peripheral blood cell counts in vivo has not yet been determined for any myeloid lineage in any species. Until the IC35 is established as the in vitro NOAEC, the ICo or perhaps IC05 should be used to estimate the NOAEL in vivo (1). Obviously, there is a pressing research need to determine the exact relationship between progenitor numbers and peripheral blood cell counts in each of the myeloid lineages and then identify the most predictive IC value to use for each level of myelosuppression in vivo. We are currently examining other compounds to determine if the IC35 consistently predicts the AUC at the NOAEL.
The Critical Importance ofIdentifying the MTC   complicated data interpretation (87,88). For example, it is common practice to compare a new compound with a reference compound of known hematotoxicity in vitro. Suppose one obtains data from such a study that shows a crossing of the concentration-response curves and it cannot be explained by differential chemical stability (Figure 3). From these data, can one predict whether humans can tolerate higher, the same, or lower levels of the investigational compound than the reference compound? In other words, should the PEL be higher or lower than the reference compound? If one knows that the IC90 is the MTC and that grade 3 to 4 neutropenia is an acceptable risk of human exposure (e.g., a cancer drug), then clearly the tolerated AUC is higher for the investigational product than for the reference compound. Assuming equivalent pharmacokinetics, the PEL can then be set higher for the new product than for the reference product. In contrast, if the IC35 is the NOAEC and neutropenia is not an acceptable consequence of exposure to this product, the opposite conclusion would be reached. The PEL for the investigational product should be set lower than the reference product because its IC35 is lower. This theoretical example illustrates how the interpretation of the in vitro data depends on the intended use of the product. For one use the PEL can be higher, whereas for another use it must be lower. This example also suggests how to choose a least hematotoxic analog from in vitro data when the analogs do not exhibit parallel curves. The least toxic analog is that with the least inhibition in an appropriate in vitro assay at the end point most relevant to the product's intended use. Such data would be impossible to interpret without these X-axis concepts.
The identical issue arises when using in vitro hematotoxicology to directly compare human toxicant sensitivity to that of the preclinical animal species in which the product is tested in vivo. In this case, the question is whether the preclinical toxicology species underor overestimates human PELs for myelosuppressive compounds. If the direct comparison results in parallel curves ( Figure 4A) then this question is easily answered. The human:animal ratio at any IC value will provide an estimate of the difference in tolerance of the compound; and, because the IC50 is the most accurately determined point on these curves, the IC50 ratios should be used. However, consider the more typical case of nonparallel curves across species ( Figure 4B). Without knowing which IC values to compare, it would be difficult to interpret these data and predict human tolerance. However, a comparison of the IC90 end point across species shows that humans can tolerate more than dogs but less than rodents. Assuming there are no pharmacokinetic differences across species, the MTD differential will be proportional to the IC9o differential. In contrast, if neutropenia is an unacceptable risk of exposure, the NOAEC will be used for comparison. Assuming the IC35 is the NOAEC, one would predict that the human PEL could be set higher than the rodent PEL by an amount proportional to the in vitro differential in the NOAECs.

Clinical Prediction Models for Prospective Evaluation andValidation Trials
Several prediction models have been developed that use the MTC and the NOAEC as predictive in vitro end points for the plasma AUC at the MTD and the NOAEL, respectively. From this estimate and knowledge of human pharmacokinetics --U---Investigational product 0 Reference toxicant 1 2 Log, exposure parameter 3 4 Figure 3. Differences between the test article and the reference compound depend on the level of inhibition at which the comparison is made. Knowing which in vitro end points are the NOAEC and the MTC makes it possible to interpret such in vitro hematotoxicology results. Note that the NOAEC is lower for the new product than the reference toxicant; in contrast, its MTC is higher. This interpretation assumes neutropenia is the dose limiting toxicity for both compounds.
Environmental Health Perspectives * Vol 106, Supplement 2 * April 1998 1 2 Log, exposure parameter  In the ideal situation, comparative toxicology data across species will map to parallel curves so that differentials in toxicant tolerance can be based on the IC50 values for two reasons: the ratio of any IC value will be the same and the IC50 is the most accurate point using conventional curve fitting methods. (B) In many cases data from multiple species must be compared across nonparallel curves that may even cross each other over the tested concentration range. In this case, knowing which in vitro end points are the NOAEC and the MTC is critical for data interpretation and clinical prediction. For example, these data predict that the MTD will be higher in rats, humans, and then dogs, assuming neutropenia is the dose-limiting toxicity and the IC90 is the MTC. In contrast, the data predict that the NOAEL will be highest in humans, followed by dogs and then rats, assuming the IC35 is the NOAEC. and plasma protein drug binding, the doses that achieve these AUCs in the plasma of exposed humans can be determined. However, the models can only be applied to compounds that are expected to show dose-limiting neutropenia. This is a critical assumption for in vitro hematotoxicology, and in vivo toxicology studies must indicate that neutrophil precursors are the dose-limiting target. If the dose-limiting toxicity of the investigational product was e.g., cardiotoxicity, determination of AUC at the MTC for CFU-GM would overestimate human tolerance and jeopardize patient safety. Some studies have used the 1C70 instead of the 1C90 as the MTC, and these two end points will be compared during some of the ongoing studies. The IC70 may be more appropriate when estimating PELs for products that will be used in patients with impaired marrow function. Prospective evaluation of these models will be critical in gaining acceptance, and Phase I clinical trials (dose-escalation trials) of antineoplastic agents provide an excellent opportunity for such an evaluation. The following models focus on the prediction of MTD; prediction of the NOAEL would simply substitute the IC35 or other NOAEC end point for the 1C90. Prediction Model 1 is the simplist model. It can be used if human pharmacokinetic parameters are unknown or cannot be determined. Model 1 has the greatest level of uncertainty because it incorporates only pharmacodynamics. It is based on the idea that neutrophil progenitors can serve as a sentinel tissue for interspecies comparisons. Large interspecies differences in toxicant disposition (AUC as a function of dose) could lead to significant errors in the predicted MTD.
Step 1: determine the MTD in an animal model; step 2: determine the toxicity differential between the animal and human dose-limiting progenitor based on IC90 values; step 3: adjust the animal MTD for the IC90 differential; and step 4: adjust the MTD again for differences in free drug concentration between the animal and human (e.g., protein binding).
Prediction Model 2 can be used when the AUC cannot be measured at the IC90 for the human progenitor but human pharmacokinetic parameters are known or can be predicted. This model assumes the marrow toxicant causes an AUC-dependent cytotoxicity. The model adjusts the plasma AUC at the MTD in the animal studies for species differentials in drug tolerance and plasma protein drug binding.
Step 1: determine the plasma AUC at the MTD in an animal model; step 2: determine the toxicity differential between the animal and human dose-limiting progenitor based on IC90 values; step 3: adjust the animalderived plasma AUC by the 1C90 differential; step 4: adjust the plasma AUC for differences in free drug concentration between species; and step 5: using human pharmacokinetic parameters or estimated parameters, calculate the dose that gives the predicted AUC.
Prediction Model 3 aims to predict the actual human MTD for the neutrophil lineage. It should be the most accurate prediction model because it incorporates both human pharmacodynamics and pharmacokinetics.
Step 1: identify the most sensitive neutrophil progenitor in human bone marrow; step 2: determine the IC90 for the dose-limiting human progenitor; step 3: derive the in vitro AUC at the IC90 by integrating the Cx t curve; step 4: translate the in vitro AUC into in vivo terms by adjusting for differences in free concentration under the two conditions of exposure (e.g., protein binding); Environmental Health Perspectives * Vol 106, Supplement 2 * April 1998 B and step 5: using human pharmacokinetic parameters or estimated parameters, calculate the dose that gives the predicted AUC. Summary Although our knowledge of how to use in vitro hematotoxicity data is in its infancy, these examples illustrate the progress that has been made by assessing exactly what end point is needed, i.e., what clinical end point should be predicted. The realization that X-axis rather than Y-axis end points are required for prediction was in a sense a breakthrough that has made it possible to propose clinical prediction models for prospective evaluation and validation (1,88,129). A second important breakthrough was our realization that predicting neutropenia actually involves the prediction of four independent parameters that, when taken together, describe clinical neutropenia: severity at nadir, time to nadir, duration of nadir, and time to recovery (84,88). Progress to date has been exclusively in the prediction of the severity of neutropenia (88,129). We are not aware that any progress that has been made in predicting the other parameters from in vitro data. Therefore, assays for CFU-GM and other granulocyte progenitors can be considered useful for investigating the cellular mechanism underlying the severity of reversible neutropenia and for determining the level of toxicant exposure that will be associated with grade 3 to 4 neutropenia. The investigation of the other parameters of neutropenia with in vitro assays should be considered exploratory research rather than established testing methodology.

Detailed Planning of in Vitro Hematotoxicology Studies
This section briefly covers some of the details that must be addressed when planning an in vitro hematotoxicology study (84,87,88).

Analytical Chemistry
There is substantial variability in the stability of different products under the conditions of the in vitro bone marrow assay, and subtle changes in the culture conditions can dramatically alter in vitro bioavailability. For example, many camptothecins show large differences in stability as a function of the species from which the culture medium serum or albumin is derived (196)(197)(198)(199)(200). These data show that one cannot assume that chemically related compounds have identical protein binding or stability in vitro and in vivo (84,88). In addition, some toxicants bind extensively to the cell culture containers (201). Therefore, the absolute Cxt curve of a test substance at the MTC and the NOAEC under the conditions of the bone marrow assay must be determined chemically to determine the AUC. Chemical analysis also provides dose confirmation. Highly misleading results can be obtained when comparing two compounds, one of which is chemically stable and the other unstable, or one that is highly soluble versus one that is partially soluble. Unstable compounds and poorly soluble compounds will appear less toxic than equally potent stable or soluble compounds simply because the former have lower in vitro bioavailability than the latter. Concluding from in vitro data that a compound is relatively nontoxic, when in fact the data reflect an artifactual rapid decomposition of compound in the culture medium, could cause serious underestimation of clinical hematotoxicity. Thus, colony inhibition data must be accompanied by sound analytical data before one can make clinical predictions, and analyte quantitation is an important aspect of the ECVAM validation study of the CFU-GM assay (1).

Duraton ofExposure
In vitro data should be obtained not only for relevant concentrations but also for relevant durations of exposure. The duration of in vitro exposure to each test substance should mimic as closely as possible the duration of in vivo exposure. For example, if in vivo exposure is multiday, brief in vitro exposures are unnecessary. Likewise, if the substance rapidly decomposes in vitro then only brief in vitro exposures should be used. In fact, prolonged in vitro exposures to unstable compounds expose the cells to potentially myelotoxic breakdown products that may not be present in vivo. The effects of decomposition on perceived toxicity become magnified as the time of compound exposure becomes much larger than the time of decomposition.
When toxicant exposure is brief, it must be decided whether to preexpose the hematopoietic cells to cytokine, stimulating their entry into the cell cycle, or to expose the hematopoietic cells immediately after isolation from the human tissue source. Toxicants that show cell cycle-dependent toxicity would be expected to show significant differences in toxicity between the two conditions. This is an important question of experimental design that needs exploration.

Endogenous (Naturally Occurrin)
Antagonists and Protein Binding ofToxicants For maximized correlation between in vitro and in vivo data, the in vitro assays must take into account physiologic levels of endogenous biochemicals that influence the toxicity of the test compound. Obvious examples of endogenous antagonists include pyrimidines and purines and their nucleosides for toxicants that target nucleoside and nucleic acid metabolism, and glutathione and other nucleophiles for alkylating agents. There are also cotoxicants such as molecular oxygen for free-radical generating compounds. The concentrations of both antagonist and cotoxicants must be as close as possible to in vivo levels.
The plasma proteins that bind xenobiotics (albumin, a1-acid glycoprotein) are another class of substances endogenous to both the in vitro and in vivo testing environments that introduce another source of error during interpretation of in vitro data. One cannot assume that a particular compound will behave identically in plasma from different species (such as from humans in vivo but from bovine or equine sources for in vitro assays), and the problem is especially significant for compounds that circulate tightly bound to carrier proteins. The levels of these proteins in the culture medium can dramatically affect the apparent potency of the test substance via changes in free (unbound) concentration of the actual toxicant (84,88,(196)(197)(198)(199)(200)(201)(202)(203). However, it is not necessary to add specific plasma proteins to the culture medium for in vitro hematotoxicity assays because results can be corrected for plasma protein binding measured in other simpler in vitro systems (1). It is best to compare several toxicants under identical culture conditions to learn about intrinsic differences in end-organ sensitivity and then correct the in vitro data for differences in protein binding determined in other assays. To facilitate in vivo extrapolation of in vitro data, the concentration-inhibition curve from bone marrow assays should express percent of inhibition as a function of free concentration of xenobiotic in the medium rather than total concentration. From these curves plus measurements of free versus total concentrations in human or animal plasma, it is possible to make in vivo predictions from in vitro data. Protoxicants and Metabolic Activation Although many toxicants are inactivated by metabolism, some compounds are Environmental Health Perspectives * Vol 106, Supplement 2 * April 1998 metabolized to toxic species. Human bone marrow stroma contains cytochrome P450s (CYP450s) that can bioactivate several toxicants, including polycyclic aromatic hydrocarbons, benzene, and cyclophosphamide (92,(204)(205)(206)(207)(208)(209)(210). However, the metabolic capacity of the marrow stroma is small compared to liver, and it cannot produce sufficient levels of metabolite to reliably detect toxicity in the in vitro assays (1,92). Several in vitro systems are available for producing human CYP450 metabolites: liver microsomes, liver slices, hepatocytes, and transgenic cell lines expressing particular CYP450 isozymes. Hepatocytes or transgenic cell lines expressing specific CYP450s can be cocultured in the in vitro assays as a physically separate cell layer (92). Alternatively, stable metabolites can be isolated and evaluated individually in the in vitro assays.
The ECVAM validation study addresses several issues related to the selection of metabolizing systems and the in vitro study of protoxicants (1). S-9 fractions, dog liver microsomes, and rat liver microsomes are commonly used to produce metabolites. However, analysis of putative hematotoxic metabolites should use species-specific sources of metabolizing enzymes, i.e., human microsomes on human bone marrow or rat microsomes on rat marrow, because of the potential for interspecies differences in bioactivation (e.g., iododoxorubicin (211)]. Xenogeneic combinations are sometimes warranted to prove the speciesspecific nature of certain toxicology problems or the lack of relevance of animal toxicology findings to humans. In addition, to exclude any contribution of metabolites to hematotoxicity, they should be tested by coexposure with the parent compound to identify any synergism or antagonism that might affect hematotoxicity.

Sources ofHuman Hematopoietic Cells
Mononuclear cells and subpopulations like CD34+ cells can be isolated from various sources: remnant marrow from femur or rib after surgeries, iliac crest aspirates from volunteers, umbilical cord blood, and peripheral blood leukocyte preparations (31,32,39,48,53,64,70,117,150,155,178,179,(212)(213)(214)(215)(216). Bone marrow-derived progenitors might be the most applicable to human toxicology, but cord blood-derived cells are more readily available for research in many countries. Pharmacologic differences, if they exist, between hematopoietic cells in adults versus cord blood may be most pronounced after exposure to cell cycle-dependent toxicants because these populations of cells show substantial differences in cell cycle status after isolation (217). One study that directly compared the chemosensitivity of human marrow and cord blood CFU-GM did not find any differences (218). In contrast, there are differences between marrow and cord blood progenitors and between mobilized and stationary progenitors in terms of autocrine cytokine stimulation, cell cycle status, and susceptibility to cytomegalovirus toxicity (150,217). Differences in cell cycle status between cord blood and bone marrow hematopoietic cells accentuate the need to determine whether preexposure to cytokines to stimulate entry into the cell cycle affects the predictive accuracy of in vitro models that use short duration of exposure.
Thoughts on the Immediate Future of in Vitro Hematotoxicology Staying Current with Experimental Hematology Experimental hematology is identifying hematopoietic cell types faster than toxicologists can determine if they are targets of toxicants. It cannot be assumed a priori that any progenitor is a target of hematotoxicants, but one cannot afford to assume that newly identified cell types are not targets and then be wrong. An efficient strategy to determine which of these cell types are targets is simply to determine which ones show altered levels in toxicantexposed animals or patients with a time course that is consistent with a direct toxicant effect. Furthermore, a new progenitor population in a preclinical species that is a bona fide toxicant target but is not found in human marrow would create a serious toxicologic problem.
If mechanism-naive assays are advantageous in regulatory science, it is important to work toward the replacement of clonogenic assays with new assays that measure the rate of production of mature blood cells by hematopoietic tissues rather than by survival of progenitors. Kinetic measurements of cell output by toxicantexposed progenitors would offer some advantages over clonogenic assays. First, the capability of the marrow to compensate for the loss of up to 30 to 40% of its progenitors via a hyperplastic response would be detectable. Second, initial rates of mature cell output by the exposed progenitor population could be analyzed sooner than colony formation (3-4 days vs 14 days), so efficiency and through-put would be increased. Third, assays that could quantify toxicant effects on the kinetics of mature cell production have the capability of detecting myelosuppression regardless of whether it is caused by cytotoxicity. Such an effect would be hard to detect and quantify using clonogenic assays (smaller colonies but no reduction in their number).

Nuanca in Daa Interpretaion
Colony-forming assays contain additional information to colony size. One of these was discussed above: the toxicant that causes smaller colonies but does not decrease their number. Consider this question: If the result of compound exposure is to inhibit the development of the erythroid component in CFU-GEMM colonies, should this be scored as inhibition of CFU-GEMM (which it probably is not) or as inhibition of an early erythroid progenitor generated by the CFU-GEMM? If the latter, would percent inhibition of BFU-E give the same estimate of myelotoxicity as the percent reduction in the CFU-GEMM that contains erythroid elements? In the design of studies to evaluate these toxicants, it may be important to limit the duration of exposure to only a few hours so that lineage-restricted progeny generated by surviving CFU-GEMM are not exposed to toxicant. Prolonged exposures would confuse toxicity to CFU-GEMM with toxicity to the progeny of CFU-GEMM.
Predicting the NOAEL raises the issue of the definition of NOAEL. Should the NOAEL be defined only by a lack of clinical symptoms and pathologic changes or should it indicate a lack of cellular and molecular changes as well? Defining the NOAEL with ever more sensitive end points, from tissue to cell to molecular levels, does not account for the compensatory mechanisms that may provide for continued health in spite of constant exposure to xenobiotics. Should PELs for a xenobiotic be based on altered gene expression in the absence of any adverse clinical effects? Rather, should it be concluded that the indicator gene, although a marker of exposure, must have some threshold yet to be exceeded and that the change does not indicate risk? The definition of NOAEL is important in regulatory science, especially as technology becomes more sensitive at detecting minute changes and more facile at expedited quests for effects. These issues must be integrated into validation studies of the NOAEC for predicting the human NOAEL.

Recommendations for Research Support
Federal agencies can take the lead in accomplishing the following goals that will promote the validation, clinical acceptance, and regulatory usefulness of in vitro hematotoxicology: * After the ECVAM study (1) validates the human CFU-GM assay for predicting the toxicant exposure level that causes grade 3 to 4 neutropenia, examine assays for the differentiated spectrum of myeloid progenitors in the marrow for predicting time to neutrophil nadir. * Fund an evaluation of possible in vitro assays and end points to predict the duration of the neutrophil nadir and/or the time to neutrophil recovery. Currently, it is much easier to investigate marrow toxicity than marrow recovery. For example, there are no in vitro assays for predicting the clinical efficacy of cytokines used to stimulate hematopoietic recovery. Research is needed to know how to predict time to recovery, which is likely related to predicting irreversible hematotoxicity as well. * Coordinate regulatoryand industrybased surveys of human bone marrow and cord blood uses and needs for in vitro hematotoxicology and propose solutions to obstacles that currently limit the use of human bone marrow in these assays, which is the actual human target tissue. * Identify the NOAEC end point in the CFU-GM assay and prove that CFU-GM but not neutrophil counts decrease by this amount in vivo at this exposure level. * Fund veterinary and clinical research to identify what progenitor stage(s) in the platelet and red blood cell lineages fluctuates in concert with peripheral blood platelet and erythrocyte levels. * Fund an evaluation of all newly identified hematopoietic cell types defined by in vitro assays and determine if acute in vivo exposure to hematotoxic xenobiotics causes a decrease in their bone marrow levels, i.e., whether these newly identified cell populations are frequent targets for xenobiotics or not. * Fund research into the replacement of clonogenic assays with new in vitro assays that quantify the rate of production of mature blood cells by toxicant-exposed hematopoietic cells.

Significance of in Vitro Hematotoxicology for Other Alternative Toxicologies
There are substantial ethical, political, financial, and regulatory pressures to replace and refine animal toxicology with alternative assays. Studies of the toxicant on its actual human target tissues would seem to be the best alternative. In vitro hematotoxicology does this, and as a result shows much promise for predicting clinical hematotoxicity, especially neutropenia caused by acute exposures. It provides an opportunity to obtain human toxicology and pharmacology information in the laboratory setting and an experimental basis for selecting an animal model for investigating clinical hematotoxicity. The in vitro assays of xenobiotic effects on human hematopoiesis can be viewed as prototypes of future in vitro toxicology assays that will reveal concepts and principles of clinical prediction. The hierarchical structure of stem cells and progenitors in the hematopoietic system likely reflects similar hierarchies in other renewing tissues of the body, such as the gastrointestinal mucosa. The emerging principles for prediction described in this paper will be applicable to toxicity in these other renewing tissues once clonogenic assays for epithelial progenitors are developed and the colony-stimulating factors are available. However, it seems unlikely that what is learned in predicting hematologic toxicity will be of much help in predicting toxicity to nonproliferative tissues such as the nervous system; other end points and principles of clinical prediction will be needed for these more troublesome toxicities.