The Baikonur Cosmodrome in Kazakhstan has over the years been an important site for rocket launching with more than two thousand launches of different rocket-carriers up to now. Today heavy equipment to the International Space Station (ISS) is transported by ‘Proton’ carriers, the propellant used for these rockets being unsymmetrical 1,1-dimethylhydrazine (1), also known as “heptyl”.

The area northeast of the Cosmodrome functions as dropping zone for burned-out rocket fuel containers of the first rocket stage separated in a height of 50 and 100 km (Proton carriers). The fuel containers at this point still contains approx. 0.6 to 4 tons of unburned 1 and about 4 tons of nitrogen oxidants [37]. Significant amounts of residual rocket fuel reach the ground, the actual amount being dependent of the season and are subsequently spread over the surface, where it either evaporates and/or penetrates into the soil [37,38]. Hence, it has been estimated that significant amounts of unburned fuel are being spread over several square kilometers of land.

In addition to the pollution with the primary pollutant 1, a series of so-called secondary pollutants being developed in soil samples polluted by 1 has recently been disclosed [5,6]. This group of compounds constitutes both transformation products that are formed directly from 1 as well as compounds that are formed in various consecutive and possibly surface catalyzed processes. In Figure 1 the the major transformation products disclosed are summarized.

In recent year there has been a special focus on compounds being persistent, bioaccumulating and toxic (PBT’s) or very persistent and very bioaccumulating (vPvB’s) [39], as such compounds obviously are of major environmental concern. Further, hazard properties for bulk chemicals are typically linked to the physical-chemical properties such as molecular weight, aqueous solubility, Henry Law constant, vapor pressure, and octanol-water partition constant and the biodegradation probability [40]. In Table 1 a selection of EPI Suite derived physico-chemical data for the 1 and its transformation products is given together with experimental data when available. The good agreement calculated and experimentally obtained values was found noted [5].

Rather high water solubility, log S_W, and correspondingly low octanol-water partition coefficients, log K_OW were found and not surprisingly low to very low Henry Law Constants, log HLC for all substances. A high migration potential for these substances was further substantiated through low water-organic carbon partition, log K_OC [5].

The majority of the compounds possess acid-base characteristics that may cause a strong affinity to mineral soil particles and thus less susceptible for biodegradation. Thus, Adushkin [41] found 1 to be very persistent in dry soils, suggesting a self-remediation period of certain soils from 1 of about 34 years.

The relatively high vapor pressures, log VP, found [4,5] were associated with an only limited evaporation from an aqueous phase, whereas evaporation from top layers of dry soils could be significant thus reducing a possible terrestrial pollution. In addition also biodegradation should be taken into account. From Table 2 is seen that all compounds apparently rapidly are degraded, the ultimate biodegradation half lives being within weeks, apart from 5 and 13. Furthermore half of the compounds were predicted to be anaerobically degradable. In Table 2 further the calculated residence times in “standard” rivers and lakes [5] (cf. Section 3.1.2) are collected.

A deeper discussion on the implications of the above figures is outside the scope of the present review and the reader is advised to consult the original papers by Carlsen et al. [4,5]. However, for completeness it should be mentioned that none of the compounds possess any significant bioaccumulation potential.

The environmental toxicity of the compounds were derived [4,5] applying the ECOSAR module of the EPI Suite leading to non-polar base line toxicity and polar acute toxicities towards fish, daphnids and green algae as summarized Table 3. Further the chronic toxicities and in certain cases the toxicities towards earthworms were predicted (data not shown here) [5].

From the figures given in Table 3 Carlsen et al. {5] concluded that apart from the primary pollutant, 1, and for the compounds 7 – 10 the investigated compounds apparently will not constitute any significant toxicity towards neither aquatic nor terrestrial organisms.

To further analyze the above data for acute toxicity a formal concept analysis was conducted in order possibly to reveal possible synergisms or antagonisms with the group of compounds [42]. In Figure 2 is displayed the line lattice diagram for ecotoxicological effects by 1 and its transformation products as derived by EcoSAR, the behind lying context table being given as Appendix 1.

Obviously the diagram contains a significant number of trivial information, like if the toxicity towards fish for a given compound < 1 mg/L it is also < 10, 100 and 1000 mg/L, respectively. However in addition to such information a series of implication sets and association rules pointed to the fact that for several of the compounds in the study toxicological effects on several species prevailed. Thus, from the FCA it was concluded [42] that for 7 compounds displaying acute toxicities towards fish at concentrations below 100 mg/L (F < 100) also displayed toxicities to daphnids below 100 mg/L (D < 100). Likewise implications were disclosed that for 5 compounds with F < 10 mg/L then D < 10 and A < 1 mg/L, for 4 compounds with F < 5 mg/L then D < 10 and A < 1 mg/L and for 5 compounds with D < 10 mg/L then F < 10 and A < 1 mg/L, respectively.

Further it was disclosed that for six out of seven compounds (86%) with D < 100 mg/L then A < 1 mg/L, for five out of six compounds (83%) with A < 1 mg/L then F < 10 and D < 10 mg/L and for four out of five compounds (80%) with A < 1, F < 10, and D < 10 mg/L then F < 5 mg/L, respectively [42].

In a further study Carlsen et al. [6] investigated the possible human health impact of 1 and its transformation products (cf. Figure 1). Thus, the probabilities for the substances to carcinogenic, mutagenic, teratogenic and/or embryotoxic were elucidated using the QSAR/QSPR software PASS (Prediction of Activity Spectra for Substances) [35], whereas absorption, distribution, metabolism and excretion (ADME) characteristics and toxicology, e.g., the probabilities for adverse organ specific health effects were disclosed using the ADME Boxes and ToxBoxes [36].

In Table 4 the results of the ADME calculations are shown. It should be noted that neither an active absorption nor any significant 1^st pass metabolism was noted for the compounds apart from 13 [6]. For a detailed discussion of the data the original study by Carlsen et al [6] should be consulted as it is outside the scope of the present review.

In the study by Carlsen et al. [6] also predicted acute toxicities of the rocket fuel and its transformation products (cf. Figure 1) were calculated (data not shown; the original reference should be consulted [6]) are compared to available experimental data.

Carlsen et al. [6] found that in some cases, e.g., in the case of 6 the predicted acute toxicities are significant overestimated, whereas in other cases, like 1 and 15 ToxBoxes apparently underestimates the toxicities. In other cases the agreement was found to be acceptable. For a more elaborate discussion the original reference [6] should be consulted.

Based on the above ADME results it was concluded [6] that the compounds apparently would move freely throughout the body and thus travelling in and out of tissues the compounds may perpetrate its biological effects. Based on calculations applying ToxBoxes (Pharma Algorithms1) the probabilities for adverse organ specific health effects (the blood, the cardiovascular and gastrointestinal systems, the kidneys, the liver and the lungs) were elucidated (Table 5). Based on these data it was concluded [6] that the most likely adverse effects are typically predicted to be in the gastrointestinal system.

Based on the above data (Table 5) the overall assessment of the adverse organ specific health effects immediately turns into a multicriteria problem as several parameters simultaneously had to be taken into account. Hence, Carlsen et al [6] advantageously applied partial order ranking [11–13,44] for the subsequent data analyses as this method allows simultaneous inclusion of several parameters. In Figure 3 the Hasse diagram constructed based on predicted adverse organ specific health effects, as derived from the ToxBoxes [36] including the gastrointestinal system (GAS), the liver (LIV) and the lungs (LUN), respectively, the more hazardous compounds being located in the top of the diagram. Thus, on a cumulative basis it was concluded [6] that compounds 4, 5 and 8 were those of major concern followed by the compounds (level 2) 1, 2, 9, 10, 12 and 13. The less hazardous compounds, 11 and 18, are found in the bottom of the diagram.

Carlsen et al. [6] further screened the 18 compounds (cf. Figure 1) for possible adverse biological effects applying the web version of the PASS software (PASS1) with the specific focus at carcinogenicity, mutagenicity, teratogenicity and embryotoxicity. In Table 6 the predicted probabilities for the studied substances being carcinogenic, mutagenic, teratogenic and embryotoxic, respectively are summarized. Only probabilities higher than 0.5 were considered.

Analogously to the above ranking of the compounds based on the adverse organ specific effects the compounds were subsequently ranked according to their probabilities ofbeing carcinogenic (CAR), mutagenic (MUT), teratogenic (TER) and embryotoxic (EMB), respectively (Figure 4).

Comparing the two figures (Figures 3 and 4) obviously some differences prevail although a series of the same compounds appear in the top levels of the diagrams. Thus, based on the PASS predictions Carlsen et al [6] found that 5 appeared as the most dangerous substances followed by the compounds (level 2) 1, 4, 6, 8 and 10, respectively. The compounds 13 and 15 – 18 are found in the bottom of the diagram as equivalent elements in agreement with the fact that these compounds all displayed probabilities less than 0.5 for the parameters studied (cf. Table 6).

Since the compounds at the same level in the diagram cannot immediately be compared Carlsen et al. [6] calculated the averaged rank of the suite of compounds studied using eqn. 4 (see Section 3) resulting in a linear rank of all compounds. In Table 7 the calculated averaged rank of the 17 compounds based on a) GAS, LIV and LUN and b) CAR, MUT, TER and EMB, respectively are given. Obviously, compounds located in the top level (level 1) in the Hasse diagrams (Figures 3 and 4) are calculated to have the top averaged ranks followed by the compounds found at the subsequent levels in the diagrams.

Subsequently an overall assessment of the human health impact by the rocket fuel 1 and its transformation products was estimated applying the Hierarchical Partial Order Ranking (HPOR) approach [6,16]. Hence, the averaged ranks given in Table 7 were adopted as so-called metaparameters [16] denoting the predicted impact according to the ToxBoxes and the PASS calculations, respectively. Subsequently a further Hasse diagram using these meta-descriptors was constructed (Figure 5) the eventual averaged rank elucidating the overall assessment of the 17 compounds with respect to their adverse human health effects are displayed in Table 8.

From Figure 5 and Table 8 Carlsen et al. [6] concluded that in addition to compounds 5 and 4, the major risk apparently would be associated with the hydrazines and the hydrazine derivatives, 1, 8, 9,10, and 12. This conclusion appeared to be parallel to the one drawn looking at the possible environmental impact (vide supra) apart from the fact that the tetrazene, 4, apparently does not appear to exhibit major risk in relation to environmental impact [5].

The here presented results (Figure 4 and 5 and Table 7 and 8) are a nice illustration of the usefulness of partial ordering methodologies in attempts to carry out assessments of, e.g., a group of xenobiotics or as studied by Carlsen et al. [6] of a group of substances consisting of a primary pollutant and a series of transformation products. Hence, through this assessment it was clearly demonstrated that some of the transformation products could lead to adverse health effects at the same or even higher level than the primary pollutant.

To further analyze the above data for adverse human health effects a formal concept analysis was conducted in order possibly to reveal possible synergisms or antagonisms with the group of compounds [42]. In Figure 6 is displayed the line lattice diagram for the probabilities of 1 and its transformation products being CAR, MUT, TER and EMB, respectively as derived by PASS, the behind lying context table being given as Appendix 2.

As in the case of the ecotoxicological data also here the diagram display a series of trivial information like, e.g. for compounds 5 the probability of being carcinogenic > 90 % (C > 90) it is of course also higher that 80 (C > 80), 70 (C > 70), 60 (C > 60), and 50 % (C >50), respectively.

However, in addition to this trivial information a series of implication sets and association rules pointed to the fact that for several of the compounds in the study a multitude of adverse human health effects prevail. In Tables 9 and 10 selected implications sets and association rules are summarized [42]. The notation like M > 70 denotes that the probability of the compounds to be mutagenic being higher that 70%. C, M, T, and E denoted carcinogenicity, mutagenicity, teratogenicity and embryotoxicity, respectively.

Although the above presented FCA studies include only a limited number of substances it nicely illustrates the possibilities to combine QSAR/QSPR generated data with formal concept analyses and thus retrieving important comprehensive information concerning the possible multitude of effects of a group of compounds.

QSAR/QSPR modeling can in the simplest form be expressed as the development of correlations between a given physico-chemical property or biological activity (endpoint), P, and a set of parameters (descriptors), D_i, that are inherent characteristics for the compounds under investigation (1) P   =   f ( D i )

The properties (endpoints), P that has been subjected to QSAR/QSPR modeling comprises physicochemical properties and biological activities in the environment as well in the human beings.

In general models that describe/calculate key properties of chemical compounds take into account three types of inherent characteristics of the molecule, i.e., structural, electronic and hydrophobic characteristics. Depending on the actual model few or many of these descriptors may be taken into account. Thus, eqn. 1 can be rewritten as (2) P   =   f ( D structural ,   D electronic ,   D hydrophobic ,   D x )   +   e

The descriptors reflecting structural characteristics may, e.g., be element of the actual composition and 3-dimensional conFiguration of the molecule, whereas descriptors reflecting the electronic characteristics may, e.g., be HOMO/LUMO energies, charge densities, dipole moment etc. The descriptors reflecting the hydrophobic characteristics are related to the distribution of the compound between a biological, hydrophobic phase, and an aqueous phase. A further, fourth type of characteristics, D_x, (cf. eqn. 2) accounts for possible underlying characteristics that may be known or unknown, such as environmental or experimental parameters as, e.g., temperature, salt content etc. The data may often be associated with a certain amount of systematic and non-quantifiable variability in combination with uncertainties. These unknown variations are expressed as “noise”. Thus, the parameter, e, account for possible noise in the system, i.e., the variation in the property that cannot be explained by the model.

In the studies presented in the present review paper a series of freely available QSAR/QSPR models has been applied. Thus, physico-chemical data, environmental persistence and environmental toxicities have been obtained applying the EPI Suite [32]. The interaction with the human organism has been elucidated through absorption, distribution, metabolism and excretion data derived by ADME Boxes [36] and the human toxicological effects by ToxBoxes [36] and by PASS (Prediction of Activity Spectra for Substances) [35].

The EPI Suite has been applied as the primary tool for generating physico-chemical endpoints [34]. This software package includes a variety of submodules to estimate, e.g., water solubility (log S_W) calculated by the submodule WSKOW, octanol-water partition (log K_OW) calculated by the submodule KOWWIN, vapor pressure (log VP) calculated by the submodule MPBPWIN, and Henry’s Law constants (log HLC) calculated by the submodule HENRY. Sorption to organic carbon was calculated using the submodule PCKOCWIN. The log K_OW values generated in this way are subsequently used to generate bioconcentration factors (log BCF) [43] calculated by the submodule BCF program. Substances with log BCF < 3.0 were regarded as non-bioaccumulating. Substances exhibiting log BCF values of > 3.0, but < 3.70 are assigned a medium bioconcentration potential whereas substances with log BCF > 3.70 were assigned a high bioconcentration potential. [34].

The theory of partial order ranking is presented elsewhere [44,54]. In brief, Partial Order Ranking is a simple principle, which a priori includes “≤” as the only mathematical relation. If a system is considered, which can be described by a series of descriptors p_i, a given site A, characterized by the descriptors p_i(A) can be compared to another site B, characterized by the descriptors p_i(B), through comparison of the single descriptors, respectively. Thus, site A will be ranked higher than site B, i.e., B ≤ A, if at least one descriptor for A is higher than the corresponding descriptor for B and no descriptor for A is lower than the corresponding descriptor for B. If, on the other hand, p_i(A) > p_i(B) for descriptor i and p_j(A) < p_j(B) for descriptor j, A and B will be denoted incomparable. Obviously, if all descriptors for A are equal to the corresponding descriptors for B, i.e., p_i(B) = p_i(A) for all i, the two sites will have identical rank and will be considered as equivalent, i.e., A = B. In mathematical terms this can be expressed as

It further follows that if A ≥ B and B ≥ C then A ≥ C. If no rank can be established between A and B these sites are denoted as incomparable, i.e., they cannot be assigned a mutual order. Therefore POR is an ideal tool to handle incommensurable attributes.

In partial order ranking – in contrast to standard multidimensional statistical analysis – neither any assumptions about linearity nor any assumptions about distribution properties are made. In this way the partial order ranking can be considered as a non-parametric method. Thus, there is no preference among the descriptors. However, due to the simple mathematics outlined above, it must be emphasized that the method a priori is rather sensitive to noise, since even minor fluctuations in the descriptor values may lead to non-comparability or reversed ordering.

A main point is that all descriptors have identical orientations, i.e., “high” and “low”. As a consequence of this, it may be necessary to multiply some descriptors by −1 in order to achieve identical directions. As an example bioaccumulation and toxicity can be mentioned. In the case of bioaccumulation, the higher the number the higher a chemical substance tends to bioaccumulate and thus the more problematic the substance, whereas in the case of toxicity, the lower the Figure the more toxic the substance. Thus, in order to secure identical directions of the two descriptors, one of them, e.g., the toxicity Figures, has to be multiplied by −1. Consequently, both in the case of bioaccumulation and in the case of toxicity higher Figures will now correspond to more problematic sites.

The graphical representation of the partial ordering is often given in a so-called Hasse diagram [55–58]. In practice the partial order rankings are performed using the WHasse software [58]. An alternative to the WHasse software is the DART (Decision Analysis by Ranking Techniques) that comprises different kinds of order ranking methods, roughly classified as total - and partial order ranking methods [59] or the PyHasse software currently being developed by R. Brüggemann [60].

Formal concept analysis (FCA) is a methodology to derive linkages between a set of objects, e.g., chemicals, and a set of associated parameters, e.g., the properties of these chemicals [17,18]. Thus, in short FCA can be as a system consisting of three parts, a context, or a triple (C,P,L), where C are the set of objects (chemicals) and P the set of parameters. L is the relation between the two sets C and P. Thus, if a chemical, c, belongs to the set C and c a parameter, p, belonging to the set P, (c,p) is said to belong to L.

The set of parameters that are associated with a given object, chemical, can be regarded as a set of binary, i.e., on/off statements. Either the chemical has a given parameter, e.g. being carcinogenic, or not.

Typically a context will be seen as arranged in matrix form with the single objects as rows and the associated parameters as columns. Hence, an “X” in this table will indicate that a given object has the given parameter (on-status) whereas an empty space indicates that this parameter is not associated with the given object (off-status). Examples of contexts are given in Table XX and YY (vide supra).

For the studies referred to in this review the software ConExp [70] was applied to generate the lattice line diagrams as well as the implication sets and association rules.