Recent years have seen a global decline in pollinator populations ( FAO, 2008 ), including surprisingly large scale losses of honey bee colonies worldwide. However, to date no consensus has been reached regarding the cause or origin of this phenomenon ( Higes et al., 2010b , 2013 ; Johnson et al., 2010 ; vanEngelsdorp & Meixer, 2010 ; Cepero et al., 2014 ). Even further, a recent analysis based on FAO data set ( Moritz & Erler, 2016 ) indicates that global scale do not show a general colony decline, while stablishing some relationship with changes on political or socioeconomic systems, as well as finding that honey trade is a dominating factor for the number of managed colonies in a country (strong import/export ratio is related with those suffering higher colony declines). Anyway all through past decade numerous reports on big colony losses are frequent and have been attributed to pesticides, pathogens, as well as climate change, landscape alteration or agricultural intensification ( Gonzalez-Varo et al., 2013 ).

One of the earliest and most studied hypotheses is related to the action of neonicotinoids and pheylpyrazole pesticides on bees ( Smirle et al., 1984 ; Halm et al., 2006 ; Johnson et al., 2010 ; Staveley et al., 2014 ). As bees encounter these pesticides while foraging for pollen or nectar, comb-stored pollen could represent a chronic source of toxicity to bees ( Thompson & Maus, 2007 ). Nevertheless, it has not yet been possible to validate this hypothesis in field conditions ( Nguyen et al., 2009 ; Higes et al., 2010b ; Cepero et al., 2014 ; Dively et al., 2015 ).

Another hypothesis suggests that different pathogens ( e.g ., viruses, microsporidia, acari, co-infections etc.) might produce a rapid decline in the adult bee population of the colony ( Higes et al., 2008 ; Cox-Foster et al., 2007 ; Blanchard et al., 2008 ; Ravoet et al., 2013 ). Several studies have directly linked parasitisation by Nosema ceranae with colony loss ( Martín-Hernández et al., 2007 ; Bacandritsos et al., 2010 ; Villa et al., 2013 ; Wolf et al., 2014 ) and indeed, infection with N. ceranae has been shown to induce sudden collapse of bee colonies under field conditions ( Higes et al., 2008 , 2009 ; Wolf et al., 2014 , Bekele et al., 2015 ). The role of Varroa destructor in the decline and death of colonies is also well established in conditions in which this mite is poorly controlled ( Rosenkranz et al., 2010 ), or it may act synergistically with other pathogens ( Cox-Foster et al., 2007 ; Ravoet et al., 2013 ). Recently, trypanosomatids were found to be a contributory factor to the colony losses in some areas ( Runckel et al., 2011 ; Ravoet et al., 2013 ) but their pathogenic effect is not yet well described and there is no consensus on the different species parasiting honeybees ( Cepero et al., 2014 ). Even nutritional stress due to habitat loss was also proposed as an alternative basis for honey bee colony collapse in the USA ( Naug, 2009 ).

As there is no consensus, the most outlined and supported hypothesis to explain these colony losses is the interaction of multiple risk factors, including those of pathogens and pesticides (Van Engelsdorp & Meixer, 2010) environmental factors ( Nguyen et al., 2009 ) and the effect of the climate change ( Le Conte & Navajas, 2008 ).

The present study focused on the region of Galicia (in northwest Spain) in an attempt to identify risk factors correlated to colony losses, which have become a serious local beekeeping problem in recent years ( Arnaiz, 2008 ).

On the basis of case definition, 53 out of the 99 samples belonged to apiaries classified by the veterinarian as affected by CL during previous year (53.5%, CI _95% =43.2-63.9; Fig. 1 ). The observed prevalence of CL difference was not statistically significant with the expected one (67%; ² = 2.90; p =0.087).

The beekeeping practices in the apiaries were very homogeneous: 95.9% of hives were Langstroth shape, 98.9% were oriented to honey production and 95% did not practice migratory beekeeping. Just 17.5% of beekeepers could be considered professional according to their operation size. None of the explanatory variables related to beekeeping practices or Varroa treatment was significantly correlated to CL ( Table 2 ; Tables S2 and S3 [suppl.] )

Even though yearly Varroa treatment is compulsory (RD608/2006) in Spain, 2% of beekeepers did not use any product to control V. destructor , while 50.5% applied two treatments per year ( Table 2 ). The most frequently applied commercial product was Apistan®, although 6.1% used homemade formulas ( e.g ., fluvalinate). The commercial products were correctly applied by 37.1% of beekeepers, and Apistan® was the product with the most correct posology (41.4%). The majority of beekeepers (98%) did not use fumagilin, an antibiotic prescribed against N. ceranae .

Palynological results of pollen samples evidenced a total of 76 types of flora detected in the brood chamber, being 94.3% (CI _95% =88.6-98.3) classified as wild vegetation ( Table 2 ). The detection of pesticides in stored pollen was rare ( Table 3 ), with fluvalinate as the most commonly identified compound (13%; CI _95% =5.9-20.3). No fipronil nor its metabolites, imidacloprid or thiametoxan were detected in any of the analyzed samples. Statistically, none of the detected compounds was correlated to CL.

Among pathogens, the most prevalent were the viruses BQCV and DWV in exterior bees (94.9%, CI _95% =88.6-98.3), and DWV in brood (81.8%, CI _95% =73.7-89.9). The most prevalent non-viral patho­gen was V. destructor in house bees (49.5%, CI _95% =39.1-59.9) with a mean infestation rate of 2.8% (SD=5.8) followed by N. ceranae in exterior bees (38.4%, CI _95% =28.3-48.5). The observed prevalence of A. woodi was higher than expected (16.2% in exterior bees; CI _95% =8.4-23.9) while P. larvae was observed in fewer bees (6.1%; CI _95% =0.9-11.3). KBV, SBV, and N. apis were rarely found, while A. apis and the viruses AIV, IAPV, SBV, APV and CPV were not detected in any of the analysed samples. While trypanosomatids were no found in brood samples, half of adult bee samples were positive (52%; CI _95% =51.5-52.5).

The presence of V. destructor in exterior and interior bees was correlated to the presence of DWV in brood samples (Pearson=0.273, p <0.01 and Pearson=0.310, p <0.01, respectively) and the presence of V. destructor in brood samples was also correlated with the presence of BQCV in brood (Pearson=0.223, p <0.05). The presence of N. ceranae and N. apis was not correlated with the presence of any other pathogen.

Only the presence of N. ceranae ( p <0.0001) and KBV (p<0.0325) were found to be significantly related to the CL ( p <0.0001). All samples with presence of N. ceranae (=38) belonged to CL positive apiaries. Positive KBV samples (=5) were few and this might be taken into account for further interpretations.

The decision tree confirmed N. ceranae as the main risk factor ( ² =53.527; p <0.0001) to CL ( Fig. 2 ). However for the samples with no presence of N. ceranae , the Varroa loads in brood ( ² =11.363; p =0.001) was the main explanatory variable yet no relation was found ( p <0.3402) from the univariate analysis. Finally the Varroa loads in worker bees ( ² =4.375; p =0.036) might explain the CL for those samples with no N. ceranae and no Varroa in brood.

The conducted study aimed to clarify the relations between CL and the presence of pesticides and the pathogen prevalence in colonies from Galicia. The study area represents approximately 4% of the total number of bee hives officially registered in Spain in 2007, but it is one of the regions in which the CL phenomenon has been discussed most actively in a wide range of forums, including beekeeper associations, as well as scientific and official institutions. This intense debate prompted the design of the present study to estimate the extent of the phenomenon and to identify risk factors involved.

Although CL is a worldwide observed phenomenon, its impact is not reflected in the official censuses in any country ( vanEngelsdorp & Meixer, 2010 ), mainly due to the rapid replacement of colonies by means of beekeeping practices. Thus, prior to this study it was difficult to estimate the real impact of colony loss in the area under study, although an earlier estimate was used as a reference ( Higes et al., 2010b ). The observed CL prevalence was 53.5% (CI _95% =43.2-63.9), lower than initially expected ( Higes et al., 2010b ). This difference might be explained with the different sample collection strategies employed in each study. Higes et al. (2010b) studied samples sent by beekeepers whose honey bee colonies were suffering from CL, potentially leading to an overestimation of the extent of CL. In the present survey, samples were collected randomly thus representing a more complete and balanced analysis of the general situation.

When analysing the possible risk factors involved in the CL phenomenon, we found that variables regarding beekeeping management such as hive type, orientation or migration activity were largely homogeneous among apiaries, and thus no link to CL was established. Only a small apiary size was associated with a lower prevalence of CL, although this association could not be established statistically probably due to the small number of apiaries of this type found.

Several pathogens were widely distributed among the apiaries studied. The ubiquity of DWV and BQCV, both with a prevalence of 94.9%, was consistent with studies performed in France ( Tentcheva et al., 2004 ), USA ( Chen et al., 2006 ) and Austria ( Berényi et al., 2006 ). As found in previous studies, we also observed correlations between each of these two viruses and V. destructor , which has been proposed to act as a mechanical viral vector ( Ball, 1989 ; Tentcheva et al ., 2004 ) or as a trigger for viral replication through its immunosuppressive activity ( Yang & Cox-Foster, 2005 ). Honey bee virus infections are usually latent, and their presence is associated with the presence of V. destructor and/or N. ceranae ( Tentcheva et al ., 2004 ; Chen & Siede, 2007 ). No other viruses, such as ABPV, SBV, AIV, IAPV or CPV were detected in the present study. IAPV was initially linked to colony losses in the USA ( Cox-Foster et al ., 2007 ) and considered a significant marker for Colony Collapse Disorder (CCD). However, in agreement with the present findings, no such association was established between this virus and the severe winter losses in other European countries ( Blanchard et al ., 2008 ; Higes et al ., 2008 ; Higes et al ., 2009 ; Garrido-Bailón et al ., 2010b ), suggesting that other pathogens play an important role in CL.

In this study, KBV was significantly linked to the presence of CL when univariate analysis was performed. Such association could not be confirmed by the multivariate analysis, and so suggesting that the role of KBV in this study might be analysed carefully. The small number of positive samples may cause a bias in the interpretation of this result.

After the viruses, V. destructor was the most common pathogen identified in the survey with prevalences ranging from 35.4% in brood samples to 49.5% in house adult bees. This high prevalence of V. destructor , a well known pathogen for which a variety of commercial treatments are available ( Rosenkranz et al ., 2010 ), may be related to the strikingly poor observance of correct treatment application. Analysis of V. destructor control strategies, which determined whether correct doses and treatments were applied, revealed a higher proportion of CL in incorrectly treated apiaries. This association could not be statistically confirmed and so we can only suggest that poor management of V. destructor might be a risk factor in the CL phenomenon. This statement is supported with the came out of the multivariate analysis that pointed out that both Varroa loads in brood and worker bees can explain some of the CL according to this statistical model.

The use of homemade formulas for V. destructor treatment in 6% of the apiaries studied is also worrying, as these probably contain unregistered acaricides or compounds that are ineffective against this pathogen. Furthermore, the identification of fluvalinate in the 13.1% of the comb-stored pollen samples suggests the undeclared use of homemade products by some beekeepers. Similar findings have been reported in other countries ( Nguyen et al ., 2009 ) and these observations are consistent with those of Johnson et al . (2010) , who wrote: “ Beekeepers searching for the primary source of pesticides contaminating bee hives need only to look in a mirror ”.

Nosema ceranae was identified in 38.4% of the apiaries sampled, a lower prevalence than that previously reported in other regions of Spain in 2005 ( Higes et al ., 2010a ). However, such figures are not surprising nowadays and according to the Brooks (1979) criteria for insect microsporidia, they represent epizootic levels of nosemosis due to N. ceranae , the agent of the nosemosis type C, so named to differentiate it from the disease caused by another honey bee microsporidia, N. apis ( Higes et al ., 2006 ). Decision tree strongly linked N. ceranae with CL as 100% apiaries that presented this microsporidium did suffer CL phenomenon. No association was found between the microsporidium species and the viruses. The most notable is the case of BQCV ( Chen & Siede, 2007 ), possibly due to the facilitation of transenteric viral infection through the cytopathogenic effects of Nosema , but it did not appear in this study.

CL in N. ceranae -positive apiaries was significantly greater than in those free of this pathogen, with an N. ceranae prevalence of 100% in the unhealthy colonies and an associated 4-fold increase in the risk of CL. This result concords with and extends the earlier findings in this country ( Higes et al ., 2010a ). While there has been debate regarding the role N. ceranae plays in CL, recent research has reached consensus that under certain conditions N. ceranae causes collapse to honey bee colonies ( Betti et al ., 2014 ; Bekele et al ., 2015 ; Cavigli et al ., 2016 ).

The observed prevalence of A. woodi (16.2%) was higher than expected and it was significantly greater in apiaries in which Varroa treatments were incorrectly applied, further emphasizing the dangers of inadequate mite control. Few studies of the last 15 years have investigated the role of A. woodi on bee colonies mortality in Europe, although McMullan & Brown (2009) qualitatively modelled the effect of tracheal mite infestation in colony death.

Some authors have linked the presence of trypanosomatids with CL ( Runckel et al ., 2011 ; Ravoet et al ., 2013 ). However they were found in half of adult bee samples, then more research is needed to clarify this implication ( Cepero et al ., 2014 ).

Palynological analysis revealed that the flora surrounding the sampled hives was diverse but mainly of wild origin rather than managed crops. No association between surrounding flora and CL was established. This observation is crucial, given that CL has been linked with nectar and pollen contamination by agropesticides used in crop management close to apiaries ( Brodschneider & Crailsheim, 2010 ; Higes et al ., 2010b ). Neither imidacloprid nor fipronil were detected in comb-stored pollen samples, despite the low detection limits of the techniques used in our study ( Nguyen et al ., 2009 ), further implying that crop pesticide contamination does not influence the CL phenomenon in the study region.

With all the information gathered, N. ceranae is the main explanatory variable associated with CL for this study. This pathogen causes chronic affection of the honey bee colonies and has been identified worldwide as primarly cause of deaths. Pathological effects of N. ceranae in Spain had been associated with the increased virulence of the Spanish strain of N. ceranae ( Huang et al ., 2012 ), or even that Apis mellifera iberiensis might be more susceptible to infection. However, the development of N. ceranae infection, bee survival and spore loads were similar in French and Spanish bee strains ( Dussaubat et al ., 2013 ), and only bees from the Netherlands appear to survive better than Spanish bees ( Van der Zee et al ., 2014 ). However, recent results showed that other beekeeping factors may have an important role in Dutch bee survival, such as the use of oxalic acid for varroa treatment ( Nannetti et al ., 2015 ) or the beneficial beekeeping management of changing queens every season ( Botias et al ., 2012 ) to avoid N. ceranae acceleration of age polyethism of young bees, causing them to exhibit behaviours typical of older bees and early death ( Lecoq et al ., 2016 ). Other issues to take into account is that the prevalence and virulence of this pathogen is higher in temperate areas like Southern Europe ( Martín-Hernández et al ., 2012 ), while its prevalence is lower in colder areas ( Fries, 2010 ). This can be explained by the ability of its spores to survive high temperatures and desiccation better than low temperatures, and the ability of N. ceranae to complete its life cycle more efficiently at high temperatures ( Fenoy et al ., 2009 ; Sanchez-Collado et al ., 2014 ).

The results of this study identify primary the presence of N. ceranae as the main explanatory variable associated with CL. However, the prevalence of V. destructor in brood cells and house bees also influences the presence of this phenomenon. The scenario found in this representative region of Spain, with a high prevalence of CL, infection with viruses, V. destructor and A. woodi , and the absence of crop pesticides, suggests that this phenomenon is related to the poor control of infectious agents rather than to toxic environmental effects.

This work is in agreement with the previous observations in a temperate climate country of the lethal effect of pathogens such as Nosema ceranae and Varroa destructor and can explain the local beekeeping big problems detected in Northern Spain region at the beginning of the XXI century.

Authors thank Gilles Budge, Neil Boonham and Katherine Perkins (Institute for Agri-Food Research and Innovation, FERA) for viruses assay and Jesús Moreno (Junta de Galicia) for all his support in the analysis of data and study design.