Mesocosm - Rio de Janeiro

#Aim

To quantify and compare the proportion of macroinvertebrates consumed by Poecilia reticulata, Poecilia vivipara, and Phalloceros harpagos under controlled mesocosm conditions.

Comparing % of Total Invertebrates consumed

To account for the large number of zero observations in the dataset, we initially fit a zero-inflated beta GLMM. The model included standard length and species as fixed effects, mesocosm identity as a random intercept, and species as a predictor of the zero-inflation probability.

##  Family: beta  ( logit )
## Formula:          X.inv ~ StandardLenght + Sp + (1 | microcosm)
## Zero inflation:         ~Sp
## Data: data_meso
## 
##       AIC       BIC    logLik -2*log(L)  df.resid 
##      59.9      78.6     -20.9      41.9        50 
## 
## Random effects:
## 
## Conditional model:
##  Groups    Name        Variance Std.Dev.
##  microcosm (Intercept) 0.4794   0.6924  
## Number of obs: 59, groups:  microcosm, 15
## 
## Dispersion parameter for beta family (): 6.02 
## 
## Conditional model:
##                Estimate Std. Error z value Pr(>|z|)   
## (Intercept)    -0.60120    0.85257  -0.705  0.48071   
## StandardLenght  0.02163    0.03159   0.685  0.49346   
## SpP. harpagos  -0.50691    0.64525  -0.786  0.43210   
## SpP. vivipara  -2.50287    0.80943  -3.092  0.00199 **
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Zero-inflation model:
##               Estimate Std. Error z value Pr(>|z|)
## (Intercept)    -0.5390     0.4756  -1.133    0.257
## SpP. harpagos   0.5390     0.6528   0.826    0.409
## SpP. vivipara   0.5390     0.6528   0.826    0.409

Zero-inflation terms were not significant (p = 0.40–1.00), and diagnostic plots showed no evidence of excess zeros. Therefore, we refit the model as a standard beta GLMM after adding a small constant (0.001) to accommodate zeros in the response.

As beta regression requires that responses fall strictly within (0,1), we added a small constant (0.001) to zero values and refit the model as a standard beta GLMM. This model provided an excellent distributional fit and was retained as the final model.

##  Family: beta  ( logit )
## Formula:          X.inv_adj ~ StandardLenght + Sp + (1 | microcosm)
## Data: data_meso
## 
##       AIC       BIC    logLik -2*log(L)  df.resid 
##    -213.1    -200.7     112.6    -225.1        53 
## 
## Random effects:
## 
## Conditional model:
##  Groups    Name        Variance Std.Dev.
##  microcosm (Intercept) 1.562    1.25    
## Number of obs: 59, groups:  microcosm, 15
## 
## Dispersion parameter for beta family (): 4.97 
## 
## Conditional model:
##                  Estimate Std. Error z value Pr(>|z|)
## (Intercept)    -1.3584508  0.9561036  -1.421    0.155
## StandardLenght  0.0004686  0.0313774   0.015    0.988
## SpP. harpagos  -0.5890332  0.8514234  -0.692    0.489
## SpP. vivipara  -1.5390830  0.9513870  -1.618    0.106

In this mesocosm experiment, species identity and body size had little influence on the proportion of total invertebrates consumed. Body size did not affect consumption (β = 0.0005 ± 0.031 SE, z = 0.02, p = 0.99).

Relative to P. reticulata (reference species), P. harpagos showed a modest but non-significant reduction in the proportion of invertebrates consumed (β = −0.589 ± 0.851, z = −0.69, p = 0.49), while P. vivipara exhibited a larger, though still non-significant, decrease (β = −1.539 ± 0.951, z = −1.62, p = 0.11).

	% of Total Invertebrates Consumed
Predictors	Estimates	p
(Intercept) — P. reticulata	-1.358 (0.956)	0.155
Standard Length (mm)	0.000 (0.031)	0.988
P. harpagos	-0.589 (0.851)	0.489
P. vivipara	-1.539 (0.951)	0.106
Random Effects
σ²	0.83
τ₀₀ _microcosm	1.56
N _microcosm	15
Observations	59

Comparing % of Chironomidae larvae consumed

Here we compare the percentage of Chironomidae consumed among the three species. We applied the same modeling approach used in the previous analysis, beginning with a zero-inflated beta GLMM and testing whether zero-inflation terms were significant.

##  Family: beta  ( logit )
## Formula:          X.chironomid ~ StandardLenght + Sp + (1 | microcosm)
## Zero inflation:                ~Sp
## Data: data_meso
## 
##       AIC       BIC    logLik -2*log(L)  df.resid 
##      56.0      74.7     -19.0      38.0        50 
## 
## Random effects:
## 
## Conditional model:
##  Groups    Name        Variance Std.Dev.
##  microcosm (Intercept) 0.02866  0.1693  
## Number of obs: 59, groups:  microcosm, 15
## 
## Dispersion parameter for beta family (): 3.81 
## 
## Conditional model:
##                 Estimate Std. Error z value Pr(>|z|)  
## (Intercept)    -0.577800   1.005667  -0.575   0.5656  
## StandardLenght -0.006201   0.042257  -0.147   0.8833  
## SpP. harpagos  -0.226575   0.473575  -0.478   0.6323  
## SpP. vivipara  -1.647577   0.727136  -2.266   0.0235 *
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Zero-inflation model:
##               Estimate Std. Error z value Pr(>|z|)  
## (Intercept)    -0.5390     0.4756  -1.133   0.2571  
## SpP. harpagos   1.1580     0.6678   1.734   0.0829 .
## SpP. vivipara   1.3863     0.6814   2.034   0.0419 *
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Zero-inflation varied among species, revealing clear contrasts between the invasive P. reticulata and native species. Native P. vivipara was significantly more likely than P. reticulata to show zero consumption of Chironomidae (β = 1.39 ± 0.68 SE, z = 2.03, p = 0.042), indicating a greater tendency to completely avoid chironomid prey, while native P. harpagos showed a similar but weaker pattern (β = 1.16 ± 0.67, z = 1.73, p = 0.083).

Conditional on feeding occurring, body size did not influence the proportion of Chironomidae consumed (β = −0.094 ± 1.04, z = −0.09, p = 0.93). Species identity strongly structured feeding intensity: native P. vivipara incorporated a significantly lower proportion of Chironomidae than the invasive P. reticulata when consumption occurred (β = −1.68 ± 0.70, z = −2.38, p = 0.017), whereas native P. harpagos did not differ from P. reticulata (β = −0.23 ± 0.48, z = −0.48, p = 0.63).

	% of Chironomidae Consumed
Predictors	Estimates	p
Count Model
(Intercept)	-0.578 (1.006)	0.566
StandardLenght	-0.006 (0.042)	0.883
Sp [P. harpagos]	-0.227 (0.474)	0.632
Sp [P. vivipara]	-1.648 (0.727)	0.023
(Intercept)	3.808 (NA)
Zero-Inflated Model
(Intercept)	-0.539 (0.476)	0.257
Sp [P. harpagos]	1.158 (0.668)	0.083
Sp [P. vivipara]	1.386 (0.681)	0.042
Random Effects
σ²	4.43
τ₀₀ _microcosm	0.03
N _microcosm	15
Observations	59

Comparing % of Culicidae larvae consumed

Here we compare the percentage of Culicidae consumed among the three species. We applied the zero-inflated beta GLMM and testing whether zero-inflation terms were significant.

##  Family: beta  ( logit )
## Formula:          X.Culicidae ~ StandardLenght + Sp + (1 | microcosm)
## Zero inflation:               ~Sp
## Data: data_meso
## 
##       AIC       BIC    logLik -2*log(L)  df.resid 
##      54.7      73.4     -18.4      36.7        50 
## 
## Random effects:
## 
## Conditional model:
##  Groups    Name        Variance  Std.Dev. 
##  microcosm (Intercept) 2.982e-10 1.727e-05
## Number of obs: 59, groups:  microcosm, 15
## 
## Dispersion parameter for beta family (): 4.46 
## 
## Conditional model:
##                Estimate Std. Error z value Pr(>|z|)  
## (Intercept)    -1.36161    0.98470  -1.383   0.1667  
## StandardLenght  0.01958    0.04198   0.466   0.6409  
## SpP. harpagos   1.01317    0.72981   1.388   0.1651  
## SpP. vivipara  -1.53854    0.92050  -1.671   0.0946 .
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Zero-inflation model:
##               Estimate Std. Error z value Pr(>|z|)  
## (Intercept)     1.0296     0.5210   1.976   0.0481 *
## SpP. harpagos   1.1676     0.9094   1.284   0.1992  
## SpP. vivipara  -0.1823     0.7138  -0.255   0.7984  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Across mesocosms, species differences in the proportion of Culicidae larvae in gut contents were modest and not statistically significant. In the conditional component of the model, body size did not influence the proportion of Culicidae consumed (β = 0.020 ± 0.042 SE, z = 0.47, p = 0.64). Relative to P. reticulata (reference species), P. harpagos tended to have a higher proportion of Culicidae in gut contents (β = 1.01 ± 0.73, z = 1.39, p = 0.17), whereas P. vivipara showed a tendency toward a lower proportion (β = −1.54 ± 0.92, z = −1.67, p = 0.095), although neither contrast was statistically significant.

Zeros in Culicidae proportions were common. In the zero-inflation component, the intercept indicated a high baseline probability of observing zero Culicidae in gut contents (β = 1.03 ± 0.52, z = 1.98, p = 0.048). However, the probability of a zero value did not differ among species (p = 0.20–0.80), suggesting that species contrasts in Culicidae proportions were driven mainly by differences in the non-zero composition of gut contents rather than species differences in the frequency of zero observations.

	% of Culicidae consumed
Predictors	Estimates	p
Count Model
(Intercept)	-1.362 (0.985)	0.167
StandardLenght	0.020 (0.042)	0.641
SpP. harpagos	1.013 (0.730)	0.165
SpP. vivipara	-1.539 (0.920)	0.095
(Intercept)	4.459 (NA)
Zero-Inflated Model
(Intercept)	1.030 (0.521)	0.048
SpP. harpagos	1.168 (0.909)	0.199
SpP. vivipara	-0.182 (0.714)	0.798
Random Effects
σ²	4.43
τ₀₀ _microcosm	0.00
N _microcosm	15
Observations	59

Mesocosm - Rio de Janeiro

Jeferson

2025-12-23

Comparing % of Total Invertebrates consumed

Comparing % of Chironomidae larvae consumed

Comparing % of Culicidae larvae consumed